Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[cascardo/linux.git] / drivers / net / ethernet / intel / ixgbevf / ixgbevf_main.c
1 /*******************************************************************************
2
3   Intel 82599 Virtual Function driver
4   Copyright(c) 1999 - 2012 Intel Corporation.
5
6   This program is free software; you can redistribute it and/or modify it
7   under the terms and conditions of the GNU General Public License,
8   version 2, as published by the Free Software Foundation.
9
10   This program is distributed in the hope it will be useful, but WITHOUT
11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13   more details.
14
15   You should have received a copy of the GNU General Public License along with
16   this program; if not, write to the Free Software Foundation, Inc.,
17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
18
19   The full GNU General Public License is included in this distribution in
20   the file called "COPYING".
21
22   Contact Information:
23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25
26 *******************************************************************************/
27
28
29 /******************************************************************************
30  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
31 ******************************************************************************/
32
33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34
35 #include <linux/types.h>
36 #include <linux/bitops.h>
37 #include <linux/module.h>
38 #include <linux/pci.h>
39 #include <linux/netdevice.h>
40 #include <linux/vmalloc.h>
41 #include <linux/string.h>
42 #include <linux/in.h>
43 #include <linux/ip.h>
44 #include <linux/tcp.h>
45 #include <linux/sctp.h>
46 #include <linux/ipv6.h>
47 #include <linux/slab.h>
48 #include <net/checksum.h>
49 #include <net/ip6_checksum.h>
50 #include <linux/ethtool.h>
51 #include <linux/if.h>
52 #include <linux/if_vlan.h>
53 #include <linux/prefetch.h>
54
55 #include "ixgbevf.h"
56
57 const char ixgbevf_driver_name[] = "ixgbevf";
58 static const char ixgbevf_driver_string[] =
59         "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
60
61 #define DRV_VERSION "2.7.12-k"
62 const char ixgbevf_driver_version[] = DRV_VERSION;
63 static char ixgbevf_copyright[] =
64         "Copyright (c) 2009 - 2012 Intel Corporation.";
65
66 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
67         [board_82599_vf] = &ixgbevf_82599_vf_info,
68         [board_X540_vf]  = &ixgbevf_X540_vf_info,
69 };
70
71 /* ixgbevf_pci_tbl - PCI Device ID Table
72  *
73  * Wildcard entries (PCI_ANY_ID) should come last
74  * Last entry must be all 0s
75  *
76  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
77  *   Class, Class Mask, private data (not used) }
78  */
79 static struct pci_device_id ixgbevf_pci_tbl[] = {
80         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF),
81         board_82599_vf},
82         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF),
83         board_X540_vf},
84
85         /* required last entry */
86         {0, }
87 };
88 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
89
90 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
91 MODULE_DESCRIPTION("Intel(R) 82599 Virtual Function Driver");
92 MODULE_LICENSE("GPL");
93 MODULE_VERSION(DRV_VERSION);
94
95 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
96 static int debug = -1;
97 module_param(debug, int, 0);
98 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
99
100 /* forward decls */
101 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
102 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
103
104 static inline void ixgbevf_release_rx_desc(struct ixgbe_hw *hw,
105                                            struct ixgbevf_ring *rx_ring,
106                                            u32 val)
107 {
108         /*
109          * Force memory writes to complete before letting h/w
110          * know there are new descriptors to fetch.  (Only
111          * applicable for weak-ordered memory model archs,
112          * such as IA-64).
113          */
114         wmb();
115         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rx_ring->reg_idx), val);
116 }
117
118 /**
119  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
120  * @adapter: pointer to adapter struct
121  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
122  * @queue: queue to map the corresponding interrupt to
123  * @msix_vector: the vector to map to the corresponding queue
124  */
125 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
126                              u8 queue, u8 msix_vector)
127 {
128         u32 ivar, index;
129         struct ixgbe_hw *hw = &adapter->hw;
130         if (direction == -1) {
131                 /* other causes */
132                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
133                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
134                 ivar &= ~0xFF;
135                 ivar |= msix_vector;
136                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
137         } else {
138                 /* tx or rx causes */
139                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
140                 index = ((16 * (queue & 1)) + (8 * direction));
141                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
142                 ivar &= ~(0xFF << index);
143                 ivar |= (msix_vector << index);
144                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
145         }
146 }
147
148 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
149                                                struct ixgbevf_tx_buffer
150                                                *tx_buffer_info)
151 {
152         if (tx_buffer_info->dma) {
153                 if (tx_buffer_info->mapped_as_page)
154                         dma_unmap_page(tx_ring->dev,
155                                        tx_buffer_info->dma,
156                                        tx_buffer_info->length,
157                                        DMA_TO_DEVICE);
158                 else
159                         dma_unmap_single(tx_ring->dev,
160                                          tx_buffer_info->dma,
161                                          tx_buffer_info->length,
162                                          DMA_TO_DEVICE);
163                 tx_buffer_info->dma = 0;
164         }
165         if (tx_buffer_info->skb) {
166                 dev_kfree_skb_any(tx_buffer_info->skb);
167                 tx_buffer_info->skb = NULL;
168         }
169         tx_buffer_info->time_stamp = 0;
170         /* tx_buffer_info must be completely set up in the transmit path */
171 }
172
173 #define IXGBE_MAX_TXD_PWR       14
174 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
175
176 /* Tx Descriptors needed, worst case */
177 #define TXD_USE_COUNT(S) DIV_ROUND_UP((S), IXGBE_MAX_DATA_PER_TXD)
178 #define DESC_NEEDED (MAX_SKB_FRAGS + 4)
179
180 static void ixgbevf_tx_timeout(struct net_device *netdev);
181
182 /**
183  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
184  * @q_vector: board private structure
185  * @tx_ring: tx ring to clean
186  **/
187 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
188                                  struct ixgbevf_ring *tx_ring)
189 {
190         struct ixgbevf_adapter *adapter = q_vector->adapter;
191         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
192         struct ixgbevf_tx_buffer *tx_buffer_info;
193         unsigned int i, eop, count = 0;
194         unsigned int total_bytes = 0, total_packets = 0;
195
196         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
197                 return true;
198
199         i = tx_ring->next_to_clean;
200         eop = tx_ring->tx_buffer_info[i].next_to_watch;
201         eop_desc = IXGBEVF_TX_DESC(tx_ring, eop);
202
203         while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
204                (count < tx_ring->count)) {
205                 bool cleaned = false;
206                 rmb(); /* read buffer_info after eop_desc */
207                 /* eop could change between read and DD-check */
208                 if (unlikely(eop != tx_ring->tx_buffer_info[i].next_to_watch))
209                         goto cont_loop;
210                 for ( ; !cleaned; count++) {
211                         struct sk_buff *skb;
212                         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
213                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
214                         cleaned = (i == eop);
215                         skb = tx_buffer_info->skb;
216
217                         if (cleaned && skb) {
218                                 unsigned int segs, bytecount;
219
220                                 /* gso_segs is currently only valid for tcp */
221                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
222                                 /* multiply data chunks by size of headers */
223                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
224                                             skb->len;
225                                 total_packets += segs;
226                                 total_bytes += bytecount;
227                         }
228
229                         ixgbevf_unmap_and_free_tx_resource(tx_ring,
230                                                            tx_buffer_info);
231
232                         tx_desc->wb.status = 0;
233
234                         i++;
235                         if (i == tx_ring->count)
236                                 i = 0;
237                 }
238
239 cont_loop:
240                 eop = tx_ring->tx_buffer_info[i].next_to_watch;
241                 eop_desc = IXGBEVF_TX_DESC(tx_ring, eop);
242         }
243
244         tx_ring->next_to_clean = i;
245
246 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
247         if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
248                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
249                 /* Make sure that anybody stopping the queue after this
250                  * sees the new next_to_clean.
251                  */
252                 smp_mb();
253                 if (__netif_subqueue_stopped(tx_ring->netdev,
254                                              tx_ring->queue_index) &&
255                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
256                         netif_wake_subqueue(tx_ring->netdev,
257                                             tx_ring->queue_index);
258                         ++adapter->restart_queue;
259                 }
260         }
261
262         u64_stats_update_begin(&tx_ring->syncp);
263         tx_ring->total_bytes += total_bytes;
264         tx_ring->total_packets += total_packets;
265         u64_stats_update_end(&tx_ring->syncp);
266         q_vector->tx.total_bytes += total_bytes;
267         q_vector->tx.total_packets += total_packets;
268
269         return count < tx_ring->count;
270 }
271
272 /**
273  * ixgbevf_receive_skb - Send a completed packet up the stack
274  * @q_vector: structure containing interrupt and ring information
275  * @skb: packet to send up
276  * @status: hardware indication of status of receive
277  * @rx_desc: rx descriptor
278  **/
279 static void ixgbevf_receive_skb(struct ixgbevf_q_vector *q_vector,
280                                 struct sk_buff *skb, u8 status,
281                                 union ixgbe_adv_rx_desc *rx_desc)
282 {
283         struct ixgbevf_adapter *adapter = q_vector->adapter;
284         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
285         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
286
287         if (is_vlan && test_bit(tag & VLAN_VID_MASK, adapter->active_vlans))
288                 __vlan_hwaccel_put_tag(skb, tag);
289
290         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
291                 napi_gro_receive(&q_vector->napi, skb);
292         else
293                 netif_rx(skb);
294 }
295
296 /**
297  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
298  * @ring: pointer to Rx descriptor ring structure
299  * @status_err: hardware indication of status of receive
300  * @skb: skb currently being received and modified
301  **/
302 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
303                                        u32 status_err, struct sk_buff *skb)
304 {
305         skb_checksum_none_assert(skb);
306
307         /* Rx csum disabled */
308         if (!(ring->netdev->features & NETIF_F_RXCSUM))
309                 return;
310
311         /* if IP and error */
312         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
313             (status_err & IXGBE_RXDADV_ERR_IPE)) {
314                 ring->hw_csum_rx_error++;
315                 return;
316         }
317
318         if (!(status_err & IXGBE_RXD_STAT_L4CS))
319                 return;
320
321         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
322                 ring->hw_csum_rx_error++;
323                 return;
324         }
325
326         /* It must be a TCP or UDP packet with a valid checksum */
327         skb->ip_summed = CHECKSUM_UNNECESSARY;
328         ring->hw_csum_rx_good++;
329 }
330
331 /**
332  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
333  * @adapter: address of board private structure
334  **/
335 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_adapter *adapter,
336                                      struct ixgbevf_ring *rx_ring,
337                                      int cleaned_count)
338 {
339         struct pci_dev *pdev = adapter->pdev;
340         union ixgbe_adv_rx_desc *rx_desc;
341         struct ixgbevf_rx_buffer *bi;
342         unsigned int i = rx_ring->next_to_use;
343
344         bi = &rx_ring->rx_buffer_info[i];
345
346         while (cleaned_count--) {
347                 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
348
349                 if (!bi->skb) {
350                         struct sk_buff *skb;
351
352                         skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
353                                                         rx_ring->rx_buf_len);
354                         if (!skb) {
355                                 adapter->alloc_rx_buff_failed++;
356                                 goto no_buffers;
357                         }
358                         bi->skb = skb;
359
360                         bi->dma = dma_map_single(&pdev->dev, skb->data,
361                                                  rx_ring->rx_buf_len,
362                                                  DMA_FROM_DEVICE);
363                         if (dma_mapping_error(&pdev->dev, bi->dma)) {
364                                 dev_kfree_skb(skb);
365                                 bi->skb = NULL;
366                                 dev_err(&pdev->dev, "RX DMA map failed\n");
367                                 break;
368                         }
369                 }
370                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
371
372                 i++;
373                 if (i == rx_ring->count)
374                         i = 0;
375                 bi = &rx_ring->rx_buffer_info[i];
376         }
377
378 no_buffers:
379         if (rx_ring->next_to_use != i) {
380                 rx_ring->next_to_use = i;
381                 ixgbevf_release_rx_desc(&adapter->hw, rx_ring, i);
382         }
383 }
384
385 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
386                                              u32 qmask)
387 {
388         struct ixgbe_hw *hw = &adapter->hw;
389
390         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
391 }
392
393 static bool ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
394                                  struct ixgbevf_ring *rx_ring,
395                                  int budget)
396 {
397         struct ixgbevf_adapter *adapter = q_vector->adapter;
398         struct pci_dev *pdev = adapter->pdev;
399         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
400         struct ixgbevf_rx_buffer *rx_buffer_info, *next_buffer;
401         struct sk_buff *skb;
402         unsigned int i;
403         u32 len, staterr;
404         int cleaned_count = 0;
405         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
406
407         i = rx_ring->next_to_clean;
408         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
409         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
410         rx_buffer_info = &rx_ring->rx_buffer_info[i];
411
412         while (staterr & IXGBE_RXD_STAT_DD) {
413                 if (!budget)
414                         break;
415                 budget--;
416
417                 rmb(); /* read descriptor and rx_buffer_info after status DD */
418                 len = le16_to_cpu(rx_desc->wb.upper.length);
419                 skb = rx_buffer_info->skb;
420                 prefetch(skb->data - NET_IP_ALIGN);
421                 rx_buffer_info->skb = NULL;
422
423                 if (rx_buffer_info->dma) {
424                         dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
425                                          rx_ring->rx_buf_len,
426                                          DMA_FROM_DEVICE);
427                         rx_buffer_info->dma = 0;
428                         skb_put(skb, len);
429                 }
430
431                 i++;
432                 if (i == rx_ring->count)
433                         i = 0;
434
435                 next_rxd = IXGBEVF_RX_DESC(rx_ring, i);
436                 prefetch(next_rxd);
437                 cleaned_count++;
438
439                 next_buffer = &rx_ring->rx_buffer_info[i];
440
441                 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
442                         skb->next = next_buffer->skb;
443                         IXGBE_CB(skb->next)->prev = skb;
444                         adapter->non_eop_descs++;
445                         goto next_desc;
446                 }
447
448                 /* we should not be chaining buffers, if we did drop the skb */
449                 if (IXGBE_CB(skb)->prev) {
450                         do {
451                                 struct sk_buff *this = skb;
452                                 skb = IXGBE_CB(skb)->prev;
453                                 dev_kfree_skb(this);
454                         } while (skb);
455                         goto next_desc;
456                 }
457
458                 /* ERR_MASK will only have valid bits if EOP set */
459                 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
460                         dev_kfree_skb_irq(skb);
461                         goto next_desc;
462                 }
463
464                 ixgbevf_rx_checksum(rx_ring, staterr, skb);
465
466                 /* probably a little skewed due to removing CRC */
467                 total_rx_bytes += skb->len;
468                 total_rx_packets++;
469
470                 /*
471                  * Work around issue of some types of VM to VM loop back
472                  * packets not getting split correctly
473                  */
474                 if (staterr & IXGBE_RXD_STAT_LB) {
475                         u32 header_fixup_len = skb_headlen(skb);
476                         if (header_fixup_len < 14)
477                                 skb_push(skb, header_fixup_len);
478                 }
479                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
480
481                 /* Workaround hardware that can't do proper VEPA multicast
482                  * source pruning.
483                  */
484                 if ((skb->pkt_type & (PACKET_BROADCAST | PACKET_MULTICAST)) &&
485                     !(compare_ether_addr(adapter->netdev->dev_addr,
486                                         eth_hdr(skb)->h_source))) {
487                         dev_kfree_skb_irq(skb);
488                         goto next_desc;
489                 }
490
491                 ixgbevf_receive_skb(q_vector, skb, staterr, rx_desc);
492
493 next_desc:
494                 rx_desc->wb.upper.status_error = 0;
495
496                 /* return some buffers to hardware, one at a time is too slow */
497                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
498                         ixgbevf_alloc_rx_buffers(adapter, rx_ring,
499                                                  cleaned_count);
500                         cleaned_count = 0;
501                 }
502
503                 /* use prefetched values */
504                 rx_desc = next_rxd;
505                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
506
507                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
508         }
509
510         rx_ring->next_to_clean = i;
511         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
512
513         if (cleaned_count)
514                 ixgbevf_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
515
516         u64_stats_update_begin(&rx_ring->syncp);
517         rx_ring->total_packets += total_rx_packets;
518         rx_ring->total_bytes += total_rx_bytes;
519         u64_stats_update_end(&rx_ring->syncp);
520         q_vector->rx.total_packets += total_rx_packets;
521         q_vector->rx.total_bytes += total_rx_bytes;
522
523         return !!budget;
524 }
525
526 /**
527  * ixgbevf_poll - NAPI polling calback
528  * @napi: napi struct with our devices info in it
529  * @budget: amount of work driver is allowed to do this pass, in packets
530  *
531  * This function will clean more than one or more rings associated with a
532  * q_vector.
533  **/
534 static int ixgbevf_poll(struct napi_struct *napi, int budget)
535 {
536         struct ixgbevf_q_vector *q_vector =
537                 container_of(napi, struct ixgbevf_q_vector, napi);
538         struct ixgbevf_adapter *adapter = q_vector->adapter;
539         struct ixgbevf_ring *ring;
540         int per_ring_budget;
541         bool clean_complete = true;
542
543         ixgbevf_for_each_ring(ring, q_vector->tx)
544                 clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
545
546         /* attempt to distribute budget to each queue fairly, but don't allow
547          * the budget to go below 1 because we'll exit polling */
548         if (q_vector->rx.count > 1)
549                 per_ring_budget = max(budget/q_vector->rx.count, 1);
550         else
551                 per_ring_budget = budget;
552
553         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
554         ixgbevf_for_each_ring(ring, q_vector->rx)
555                 clean_complete &= ixgbevf_clean_rx_irq(q_vector, ring,
556                                                        per_ring_budget);
557         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
558
559         /* If all work not completed, return budget and keep polling */
560         if (!clean_complete)
561                 return budget;
562         /* all work done, exit the polling mode */
563         napi_complete(napi);
564         if (adapter->rx_itr_setting & 1)
565                 ixgbevf_set_itr(q_vector);
566         if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
567                 ixgbevf_irq_enable_queues(adapter,
568                                           1 << q_vector->v_idx);
569
570         return 0;
571 }
572
573 /**
574  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
575  * @q_vector: structure containing interrupt and ring information
576  */
577 static void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
578 {
579         struct ixgbevf_adapter *adapter = q_vector->adapter;
580         struct ixgbe_hw *hw = &adapter->hw;
581         int v_idx = q_vector->v_idx;
582         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
583
584         /*
585          * set the WDIS bit to not clear the timer bits and cause an
586          * immediate assertion of the interrupt
587          */
588         itr_reg |= IXGBE_EITR_CNT_WDIS;
589
590         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
591 }
592
593 /**
594  * ixgbevf_configure_msix - Configure MSI-X hardware
595  * @adapter: board private structure
596  *
597  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
598  * interrupts.
599  **/
600 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
601 {
602         struct ixgbevf_q_vector *q_vector;
603         int q_vectors, v_idx;
604
605         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
606         adapter->eims_enable_mask = 0;
607
608         /*
609          * Populate the IVAR table and set the ITR values to the
610          * corresponding register.
611          */
612         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
613                 struct ixgbevf_ring *ring;
614                 q_vector = adapter->q_vector[v_idx];
615
616                 ixgbevf_for_each_ring(ring, q_vector->rx)
617                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
618
619                 ixgbevf_for_each_ring(ring, q_vector->tx)
620                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
621
622                 if (q_vector->tx.ring && !q_vector->rx.ring) {
623                         /* tx only vector */
624                         if (adapter->tx_itr_setting == 1)
625                                 q_vector->itr = IXGBE_10K_ITR;
626                         else
627                                 q_vector->itr = adapter->tx_itr_setting;
628                 } else {
629                         /* rx or rx/tx vector */
630                         if (adapter->rx_itr_setting == 1)
631                                 q_vector->itr = IXGBE_20K_ITR;
632                         else
633                                 q_vector->itr = adapter->rx_itr_setting;
634                 }
635
636                 /* add q_vector eims value to global eims_enable_mask */
637                 adapter->eims_enable_mask |= 1 << v_idx;
638
639                 ixgbevf_write_eitr(q_vector);
640         }
641
642         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
643         /* setup eims_other and add value to global eims_enable_mask */
644         adapter->eims_other = 1 << v_idx;
645         adapter->eims_enable_mask |= adapter->eims_other;
646 }
647
648 enum latency_range {
649         lowest_latency = 0,
650         low_latency = 1,
651         bulk_latency = 2,
652         latency_invalid = 255
653 };
654
655 /**
656  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
657  * @q_vector: structure containing interrupt and ring information
658  * @ring_container: structure containing ring performance data
659  *
660  *      Stores a new ITR value based on packets and byte
661  *      counts during the last interrupt.  The advantage of per interrupt
662  *      computation is faster updates and more accurate ITR for the current
663  *      traffic pattern.  Constants in this function were computed
664  *      based on theoretical maximum wire speed and thresholds were set based
665  *      on testing data as well as attempting to minimize response time
666  *      while increasing bulk throughput.
667  **/
668 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
669                                struct ixgbevf_ring_container *ring_container)
670 {
671         int bytes = ring_container->total_bytes;
672         int packets = ring_container->total_packets;
673         u32 timepassed_us;
674         u64 bytes_perint;
675         u8 itr_setting = ring_container->itr;
676
677         if (packets == 0)
678                 return;
679
680         /* simple throttlerate management
681          *    0-20MB/s lowest (100000 ints/s)
682          *   20-100MB/s low   (20000 ints/s)
683          *  100-1249MB/s bulk (8000 ints/s)
684          */
685         /* what was last interrupt timeslice? */
686         timepassed_us = q_vector->itr >> 2;
687         bytes_perint = bytes / timepassed_us; /* bytes/usec */
688
689         switch (itr_setting) {
690         case lowest_latency:
691                 if (bytes_perint > 10)
692                         itr_setting = low_latency;
693                 break;
694         case low_latency:
695                 if (bytes_perint > 20)
696                         itr_setting = bulk_latency;
697                 else if (bytes_perint <= 10)
698                         itr_setting = lowest_latency;
699                 break;
700         case bulk_latency:
701                 if (bytes_perint <= 20)
702                         itr_setting = low_latency;
703                 break;
704         }
705
706         /* clear work counters since we have the values we need */
707         ring_container->total_bytes = 0;
708         ring_container->total_packets = 0;
709
710         /* write updated itr to ring container */
711         ring_container->itr = itr_setting;
712 }
713
714 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
715 {
716         u32 new_itr = q_vector->itr;
717         u8 current_itr;
718
719         ixgbevf_update_itr(q_vector, &q_vector->tx);
720         ixgbevf_update_itr(q_vector, &q_vector->rx);
721
722         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
723
724         switch (current_itr) {
725         /* counts and packets in update_itr are dependent on these numbers */
726         case lowest_latency:
727                 new_itr = IXGBE_100K_ITR;
728                 break;
729         case low_latency:
730                 new_itr = IXGBE_20K_ITR;
731                 break;
732         case bulk_latency:
733         default:
734                 new_itr = IXGBE_8K_ITR;
735                 break;
736         }
737
738         if (new_itr != q_vector->itr) {
739                 /* do an exponential smoothing */
740                 new_itr = (10 * new_itr * q_vector->itr) /
741                           ((9 * new_itr) + q_vector->itr);
742
743                 /* save the algorithm value here */
744                 q_vector->itr = new_itr;
745
746                 ixgbevf_write_eitr(q_vector);
747         }
748 }
749
750 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
751 {
752         struct ixgbevf_adapter *adapter = data;
753         struct pci_dev *pdev = adapter->pdev;
754         struct ixgbe_hw *hw = &adapter->hw;
755         u32 msg;
756         bool got_ack = false;
757
758         hw->mac.get_link_status = 1;
759         if (!hw->mbx.ops.check_for_ack(hw))
760                 got_ack = true;
761
762         if (!hw->mbx.ops.check_for_msg(hw)) {
763                 hw->mbx.ops.read(hw, &msg, 1);
764
765                 if ((msg & IXGBE_MBVFICR_VFREQ_MASK) == IXGBE_PF_CONTROL_MSG) {
766                         mod_timer(&adapter->watchdog_timer,
767                                   round_jiffies(jiffies + 1));
768                         adapter->link_up = false;
769                 }
770
771                 if (msg & IXGBE_VT_MSGTYPE_NACK)
772                         dev_info(&pdev->dev,
773                                  "Last Request of type %2.2x to PF Nacked\n",
774                                  msg & 0xFF);
775                 hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFSTS;
776         }
777
778         /* checking for the ack clears the PFACK bit.  Place
779          * it back in the v2p_mailbox cache so that anyone
780          * polling for an ack will not miss it
781          */
782         if (got_ack)
783                 hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFACK;
784
785         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
786
787         return IRQ_HANDLED;
788 }
789
790 /**
791  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
792  * @irq: unused
793  * @data: pointer to our q_vector struct for this interrupt vector
794  **/
795 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
796 {
797         struct ixgbevf_q_vector *q_vector = data;
798
799         /* EIAM disabled interrupts (on this vector) for us */
800         if (q_vector->rx.ring || q_vector->tx.ring)
801                 napi_schedule(&q_vector->napi);
802
803         return IRQ_HANDLED;
804 }
805
806 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
807                                      int r_idx)
808 {
809         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
810
811         a->rx_ring[r_idx].next = q_vector->rx.ring;
812         q_vector->rx.ring = &a->rx_ring[r_idx];
813         q_vector->rx.count++;
814 }
815
816 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
817                                      int t_idx)
818 {
819         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
820
821         a->tx_ring[t_idx].next = q_vector->tx.ring;
822         q_vector->tx.ring = &a->tx_ring[t_idx];
823         q_vector->tx.count++;
824 }
825
826 /**
827  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
828  * @adapter: board private structure to initialize
829  *
830  * This function maps descriptor rings to the queue-specific vectors
831  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
832  * one vector per ring/queue, but on a constrained vector budget, we
833  * group the rings as "efficiently" as possible.  You would add new
834  * mapping configurations in here.
835  **/
836 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
837 {
838         int q_vectors;
839         int v_start = 0;
840         int rxr_idx = 0, txr_idx = 0;
841         int rxr_remaining = adapter->num_rx_queues;
842         int txr_remaining = adapter->num_tx_queues;
843         int i, j;
844         int rqpv, tqpv;
845         int err = 0;
846
847         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
848
849         /*
850          * The ideal configuration...
851          * We have enough vectors to map one per queue.
852          */
853         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
854                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
855                         map_vector_to_rxq(adapter, v_start, rxr_idx);
856
857                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
858                         map_vector_to_txq(adapter, v_start, txr_idx);
859                 goto out;
860         }
861
862         /*
863          * If we don't have enough vectors for a 1-to-1
864          * mapping, we'll have to group them so there are
865          * multiple queues per vector.
866          */
867         /* Re-adjusting *qpv takes care of the remainder. */
868         for (i = v_start; i < q_vectors; i++) {
869                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
870                 for (j = 0; j < rqpv; j++) {
871                         map_vector_to_rxq(adapter, i, rxr_idx);
872                         rxr_idx++;
873                         rxr_remaining--;
874                 }
875         }
876         for (i = v_start; i < q_vectors; i++) {
877                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
878                 for (j = 0; j < tqpv; j++) {
879                         map_vector_to_txq(adapter, i, txr_idx);
880                         txr_idx++;
881                         txr_remaining--;
882                 }
883         }
884
885 out:
886         return err;
887 }
888
889 /**
890  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
891  * @adapter: board private structure
892  *
893  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
894  * interrupts from the kernel.
895  **/
896 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
897 {
898         struct net_device *netdev = adapter->netdev;
899         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
900         int vector, err;
901         int ri = 0, ti = 0;
902
903         for (vector = 0; vector < q_vectors; vector++) {
904                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
905                 struct msix_entry *entry = &adapter->msix_entries[vector];
906
907                 if (q_vector->tx.ring && q_vector->rx.ring) {
908                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
909                                  "%s-%s-%d", netdev->name, "TxRx", ri++);
910                         ti++;
911                 } else if (q_vector->rx.ring) {
912                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
913                                  "%s-%s-%d", netdev->name, "rx", ri++);
914                 } else if (q_vector->tx.ring) {
915                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
916                                  "%s-%s-%d", netdev->name, "tx", ti++);
917                 } else {
918                         /* skip this unused q_vector */
919                         continue;
920                 }
921                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
922                                   q_vector->name, q_vector);
923                 if (err) {
924                         hw_dbg(&adapter->hw,
925                                "request_irq failed for MSIX interrupt "
926                                "Error: %d\n", err);
927                         goto free_queue_irqs;
928                 }
929         }
930
931         err = request_irq(adapter->msix_entries[vector].vector,
932                           &ixgbevf_msix_other, 0, netdev->name, adapter);
933         if (err) {
934                 hw_dbg(&adapter->hw,
935                        "request_irq for msix_other failed: %d\n", err);
936                 goto free_queue_irqs;
937         }
938
939         return 0;
940
941 free_queue_irqs:
942         while (vector) {
943                 vector--;
944                 free_irq(adapter->msix_entries[vector].vector,
945                          adapter->q_vector[vector]);
946         }
947         pci_disable_msix(adapter->pdev);
948         kfree(adapter->msix_entries);
949         adapter->msix_entries = NULL;
950         return err;
951 }
952
953 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
954 {
955         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
956
957         for (i = 0; i < q_vectors; i++) {
958                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
959                 q_vector->rx.ring = NULL;
960                 q_vector->tx.ring = NULL;
961                 q_vector->rx.count = 0;
962                 q_vector->tx.count = 0;
963         }
964 }
965
966 /**
967  * ixgbevf_request_irq - initialize interrupts
968  * @adapter: board private structure
969  *
970  * Attempts to configure interrupts using the best available
971  * capabilities of the hardware and kernel.
972  **/
973 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
974 {
975         int err = 0;
976
977         err = ixgbevf_request_msix_irqs(adapter);
978
979         if (err)
980                 hw_dbg(&adapter->hw,
981                        "request_irq failed, Error %d\n", err);
982
983         return err;
984 }
985
986 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
987 {
988         int i, q_vectors;
989
990         q_vectors = adapter->num_msix_vectors;
991         i = q_vectors - 1;
992
993         free_irq(adapter->msix_entries[i].vector, adapter);
994         i--;
995
996         for (; i >= 0; i--) {
997                 /* free only the irqs that were actually requested */
998                 if (!adapter->q_vector[i]->rx.ring &&
999                     !adapter->q_vector[i]->tx.ring)
1000                         continue;
1001
1002                 free_irq(adapter->msix_entries[i].vector,
1003                          adapter->q_vector[i]);
1004         }
1005
1006         ixgbevf_reset_q_vectors(adapter);
1007 }
1008
1009 /**
1010  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1011  * @adapter: board private structure
1012  **/
1013 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1014 {
1015         struct ixgbe_hw *hw = &adapter->hw;
1016         int i;
1017
1018         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1019         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1020         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1021
1022         IXGBE_WRITE_FLUSH(hw);
1023
1024         for (i = 0; i < adapter->num_msix_vectors; i++)
1025                 synchronize_irq(adapter->msix_entries[i].vector);
1026 }
1027
1028 /**
1029  * ixgbevf_irq_enable - Enable default interrupt generation settings
1030  * @adapter: board private structure
1031  **/
1032 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1033 {
1034         struct ixgbe_hw *hw = &adapter->hw;
1035
1036         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1037         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1038         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1039 }
1040
1041 /**
1042  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1043  * @adapter: board private structure
1044  *
1045  * Configure the Tx unit of the MAC after a reset.
1046  **/
1047 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1048 {
1049         u64 tdba;
1050         struct ixgbe_hw *hw = &adapter->hw;
1051         u32 i, j, tdlen, txctrl;
1052
1053         /* Setup the HW Tx Head and Tail descriptor pointers */
1054         for (i = 0; i < adapter->num_tx_queues; i++) {
1055                 struct ixgbevf_ring *ring = &adapter->tx_ring[i];
1056                 j = ring->reg_idx;
1057                 tdba = ring->dma;
1058                 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1059                 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1060                                 (tdba & DMA_BIT_MASK(32)));
1061                 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1062                 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j), tdlen);
1063                 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(j), 0);
1064                 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(j), 0);
1065                 adapter->tx_ring[i].head = IXGBE_VFTDH(j);
1066                 adapter->tx_ring[i].tail = IXGBE_VFTDT(j);
1067                 /* Disable Tx Head Writeback RO bit, since this hoses
1068                  * bookkeeping if things aren't delivered in order.
1069                  */
1070                 txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1071                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1072                 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1073         }
1074 }
1075
1076 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1077
1078 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1079 {
1080         struct ixgbevf_ring *rx_ring;
1081         struct ixgbe_hw *hw = &adapter->hw;
1082         u32 srrctl;
1083
1084         rx_ring = &adapter->rx_ring[index];
1085
1086         srrctl = IXGBE_SRRCTL_DROP_EN;
1087
1088         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1089
1090         srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
1091                   IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1092
1093         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1094 }
1095
1096 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter)
1097 {
1098         struct ixgbe_hw *hw = &adapter->hw;
1099         struct net_device *netdev = adapter->netdev;
1100         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1101         int i;
1102         u16 rx_buf_len;
1103
1104         /* notify the PF of our intent to use this size of frame */
1105         ixgbevf_rlpml_set_vf(hw, max_frame);
1106
1107         /* PF will allow an extra 4 bytes past for vlan tagged frames */
1108         max_frame += VLAN_HLEN;
1109
1110         /*
1111          * Allocate buffer sizes that fit well into 32K and
1112          * take into account max frame size of 9.5K
1113          */
1114         if ((hw->mac.type == ixgbe_mac_X540_vf) &&
1115             (max_frame <= MAXIMUM_ETHERNET_VLAN_SIZE))
1116                 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1117         else if (max_frame <= IXGBEVF_RXBUFFER_2K)
1118                 rx_buf_len = IXGBEVF_RXBUFFER_2K;
1119         else if (max_frame <= IXGBEVF_RXBUFFER_4K)
1120                 rx_buf_len = IXGBEVF_RXBUFFER_4K;
1121         else if (max_frame <= IXGBEVF_RXBUFFER_8K)
1122                 rx_buf_len = IXGBEVF_RXBUFFER_8K;
1123         else
1124                 rx_buf_len = IXGBEVF_RXBUFFER_10K;
1125
1126         for (i = 0; i < adapter->num_rx_queues; i++)
1127                 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1128 }
1129
1130 /**
1131  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1132  * @adapter: board private structure
1133  *
1134  * Configure the Rx unit of the MAC after a reset.
1135  **/
1136 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1137 {
1138         u64 rdba;
1139         struct ixgbe_hw *hw = &adapter->hw;
1140         int i, j;
1141         u32 rdlen;
1142
1143         /* PSRTYPE must be initialized in 82599 */
1144         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, 0);
1145
1146         /* set_rx_buffer_len must be called before ring initialization */
1147         ixgbevf_set_rx_buffer_len(adapter);
1148
1149         rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1150         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1151          * the Base and Length of the Rx Descriptor Ring */
1152         for (i = 0; i < adapter->num_rx_queues; i++) {
1153                 rdba = adapter->rx_ring[i].dma;
1154                 j = adapter->rx_ring[i].reg_idx;
1155                 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1156                                 (rdba & DMA_BIT_MASK(32)));
1157                 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1158                 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j), rdlen);
1159                 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(j), 0);
1160                 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(j), 0);
1161                 adapter->rx_ring[i].head = IXGBE_VFRDH(j);
1162                 adapter->rx_ring[i].tail = IXGBE_VFRDT(j);
1163
1164                 ixgbevf_configure_srrctl(adapter, j);
1165         }
1166 }
1167
1168 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1169 {
1170         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1171         struct ixgbe_hw *hw = &adapter->hw;
1172         int err;
1173
1174         spin_lock_bh(&adapter->mbx_lock);
1175
1176         /* add VID to filter table */
1177         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1178
1179         spin_unlock_bh(&adapter->mbx_lock);
1180
1181         /* translate error return types so error makes sense */
1182         if (err == IXGBE_ERR_MBX)
1183                 return -EIO;
1184
1185         if (err == IXGBE_ERR_INVALID_ARGUMENT)
1186                 return -EACCES;
1187
1188         set_bit(vid, adapter->active_vlans);
1189
1190         return err;
1191 }
1192
1193 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1194 {
1195         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1196         struct ixgbe_hw *hw = &adapter->hw;
1197         int err = -EOPNOTSUPP;
1198
1199         spin_lock_bh(&adapter->mbx_lock);
1200
1201         /* remove VID from filter table */
1202         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1203
1204         spin_unlock_bh(&adapter->mbx_lock);
1205
1206         clear_bit(vid, adapter->active_vlans);
1207
1208         return err;
1209 }
1210
1211 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1212 {
1213         u16 vid;
1214
1215         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1216                 ixgbevf_vlan_rx_add_vid(adapter->netdev, vid);
1217 }
1218
1219 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1220 {
1221         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1222         struct ixgbe_hw *hw = &adapter->hw;
1223         int count = 0;
1224
1225         if ((netdev_uc_count(netdev)) > 10) {
1226                 pr_err("Too many unicast filters - No Space\n");
1227                 return -ENOSPC;
1228         }
1229
1230         if (!netdev_uc_empty(netdev)) {
1231                 struct netdev_hw_addr *ha;
1232                 netdev_for_each_uc_addr(ha, netdev) {
1233                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1234                         udelay(200);
1235                 }
1236         } else {
1237                 /*
1238                  * If the list is empty then send message to PF driver to
1239                  * clear all macvlans on this VF.
1240                  */
1241                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1242         }
1243
1244         return count;
1245 }
1246
1247 /**
1248  * ixgbevf_set_rx_mode - Multicast and unicast set
1249  * @netdev: network interface device structure
1250  *
1251  * The set_rx_method entry point is called whenever the multicast address
1252  * list, unicast address list or the network interface flags are updated.
1253  * This routine is responsible for configuring the hardware for proper
1254  * multicast mode and configuring requested unicast filters.
1255  **/
1256 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1257 {
1258         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1259         struct ixgbe_hw *hw = &adapter->hw;
1260
1261         spin_lock_bh(&adapter->mbx_lock);
1262
1263         /* reprogram multicast list */
1264         hw->mac.ops.update_mc_addr_list(hw, netdev);
1265
1266         ixgbevf_write_uc_addr_list(netdev);
1267
1268         spin_unlock_bh(&adapter->mbx_lock);
1269 }
1270
1271 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1272 {
1273         int q_idx;
1274         struct ixgbevf_q_vector *q_vector;
1275         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1276
1277         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1278                 q_vector = adapter->q_vector[q_idx];
1279                 napi_enable(&q_vector->napi);
1280         }
1281 }
1282
1283 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1284 {
1285         int q_idx;
1286         struct ixgbevf_q_vector *q_vector;
1287         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1288
1289         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1290                 q_vector = adapter->q_vector[q_idx];
1291                 napi_disable(&q_vector->napi);
1292         }
1293 }
1294
1295 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1296 {
1297         struct net_device *netdev = adapter->netdev;
1298         int i;
1299
1300         ixgbevf_set_rx_mode(netdev);
1301
1302         ixgbevf_restore_vlan(adapter);
1303
1304         ixgbevf_configure_tx(adapter);
1305         ixgbevf_configure_rx(adapter);
1306         for (i = 0; i < adapter->num_rx_queues; i++) {
1307                 struct ixgbevf_ring *ring = &adapter->rx_ring[i];
1308                 ixgbevf_alloc_rx_buffers(adapter, ring,
1309                                          IXGBE_DESC_UNUSED(ring));
1310         }
1311 }
1312
1313 #define IXGBE_MAX_RX_DESC_POLL 10
1314 static inline void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1315                                                 int rxr)
1316 {
1317         struct ixgbe_hw *hw = &adapter->hw;
1318         int j = adapter->rx_ring[rxr].reg_idx;
1319         int k;
1320
1321         for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
1322                 if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
1323                         break;
1324                 else
1325                         msleep(1);
1326         }
1327         if (k >= IXGBE_MAX_RX_DESC_POLL) {
1328                 hw_dbg(hw, "RXDCTL.ENABLE on Rx queue %d "
1329                        "not set within the polling period\n", rxr);
1330         }
1331
1332         ixgbevf_release_rx_desc(hw, &adapter->rx_ring[rxr],
1333                                 adapter->rx_ring[rxr].count - 1);
1334 }
1335
1336 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1337 {
1338         /* Only save pre-reset stats if there are some */
1339         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1340                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1341                         adapter->stats.base_vfgprc;
1342                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1343                         adapter->stats.base_vfgptc;
1344                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1345                         adapter->stats.base_vfgorc;
1346                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1347                         adapter->stats.base_vfgotc;
1348                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1349                         adapter->stats.base_vfmprc;
1350         }
1351 }
1352
1353 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1354 {
1355         struct ixgbe_hw *hw = &adapter->hw;
1356
1357         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1358         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1359         adapter->stats.last_vfgorc |=
1360                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1361         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1362         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1363         adapter->stats.last_vfgotc |=
1364                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1365         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1366
1367         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1368         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1369         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1370         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1371         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
1372 }
1373
1374 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
1375 {
1376         struct ixgbe_hw *hw = &adapter->hw;
1377         int api[] = { ixgbe_mbox_api_11,
1378                       ixgbe_mbox_api_10,
1379                       ixgbe_mbox_api_unknown };
1380         int err = 0, idx = 0;
1381
1382         spin_lock_bh(&adapter->mbx_lock);
1383
1384         while (api[idx] != ixgbe_mbox_api_unknown) {
1385                 err = ixgbevf_negotiate_api_version(hw, api[idx]);
1386                 if (!err)
1387                         break;
1388                 idx++;
1389         }
1390
1391         spin_unlock_bh(&adapter->mbx_lock);
1392 }
1393
1394 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
1395 {
1396         struct net_device *netdev = adapter->netdev;
1397         struct ixgbe_hw *hw = &adapter->hw;
1398         int i, j = 0;
1399         int num_rx_rings = adapter->num_rx_queues;
1400         u32 txdctl, rxdctl;
1401
1402         for (i = 0; i < adapter->num_tx_queues; i++) {
1403                 j = adapter->tx_ring[i].reg_idx;
1404                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1405                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
1406                 txdctl |= (8 << 16);
1407                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1408         }
1409
1410         for (i = 0; i < adapter->num_tx_queues; i++) {
1411                 j = adapter->tx_ring[i].reg_idx;
1412                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1413                 txdctl |= IXGBE_TXDCTL_ENABLE;
1414                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1415         }
1416
1417         for (i = 0; i < num_rx_rings; i++) {
1418                 j = adapter->rx_ring[i].reg_idx;
1419                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1420                 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1421                 if (hw->mac.type == ixgbe_mac_X540_vf) {
1422                         rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
1423                         rxdctl |= ((netdev->mtu + ETH_HLEN + ETH_FCS_LEN) |
1424                                    IXGBE_RXDCTL_RLPML_EN);
1425                 }
1426                 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1427                 ixgbevf_rx_desc_queue_enable(adapter, i);
1428         }
1429
1430         ixgbevf_configure_msix(adapter);
1431
1432         spin_lock_bh(&adapter->mbx_lock);
1433
1434         if (is_valid_ether_addr(hw->mac.addr))
1435                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
1436         else
1437                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
1438
1439         spin_unlock_bh(&adapter->mbx_lock);
1440
1441         clear_bit(__IXGBEVF_DOWN, &adapter->state);
1442         ixgbevf_napi_enable_all(adapter);
1443
1444         /* enable transmits */
1445         netif_tx_start_all_queues(netdev);
1446
1447         ixgbevf_save_reset_stats(adapter);
1448         ixgbevf_init_last_counter_stats(adapter);
1449
1450         hw->mac.get_link_status = 1;
1451         mod_timer(&adapter->watchdog_timer, jiffies);
1452 }
1453
1454 static int ixgbevf_reset_queues(struct ixgbevf_adapter *adapter)
1455 {
1456         struct ixgbe_hw *hw = &adapter->hw;
1457         struct ixgbevf_ring *rx_ring;
1458         unsigned int def_q = 0;
1459         unsigned int num_tcs = 0;
1460         unsigned int num_rx_queues = 1;
1461         int err, i;
1462
1463         spin_lock_bh(&adapter->mbx_lock);
1464
1465         /* fetch queue configuration from the PF */
1466         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1467
1468         spin_unlock_bh(&adapter->mbx_lock);
1469
1470         if (err)
1471                 return err;
1472
1473         if (num_tcs > 1) {
1474                 /* update default Tx ring register index */
1475                 adapter->tx_ring[0].reg_idx = def_q;
1476
1477                 /* we need as many queues as traffic classes */
1478                 num_rx_queues = num_tcs;
1479         }
1480
1481         /* nothing to do if we have the correct number of queues */
1482         if (adapter->num_rx_queues == num_rx_queues)
1483                 return 0;
1484
1485         /* allocate new rings */
1486         rx_ring = kcalloc(num_rx_queues,
1487                           sizeof(struct ixgbevf_ring), GFP_KERNEL);
1488         if (!rx_ring)
1489                 return -ENOMEM;
1490
1491         /* setup ring fields */
1492         for (i = 0; i < num_rx_queues; i++) {
1493                 rx_ring[i].count = adapter->rx_ring_count;
1494                 rx_ring[i].queue_index = i;
1495                 rx_ring[i].reg_idx = i;
1496                 rx_ring[i].dev = &adapter->pdev->dev;
1497                 rx_ring[i].netdev = adapter->netdev;
1498
1499                 /* allocate resources on the ring */
1500                 err = ixgbevf_setup_rx_resources(adapter, &rx_ring[i]);
1501                 if (err) {
1502                         while (i) {
1503                                 i--;
1504                                 ixgbevf_free_rx_resources(adapter, &rx_ring[i]);
1505                         }
1506                         kfree(rx_ring);
1507                         return err;
1508                 }
1509         }
1510
1511         /* free the existing rings and queues */
1512         ixgbevf_free_all_rx_resources(adapter);
1513         adapter->num_rx_queues = 0;
1514         kfree(adapter->rx_ring);
1515
1516         /* move new rings into position on the adapter struct */
1517         adapter->rx_ring = rx_ring;
1518         adapter->num_rx_queues = num_rx_queues;
1519
1520         /* reset ring to vector mapping */
1521         ixgbevf_reset_q_vectors(adapter);
1522         ixgbevf_map_rings_to_vectors(adapter);
1523
1524         return 0;
1525 }
1526
1527 void ixgbevf_up(struct ixgbevf_adapter *adapter)
1528 {
1529         struct ixgbe_hw *hw = &adapter->hw;
1530
1531         ixgbevf_negotiate_api(adapter);
1532
1533         ixgbevf_reset_queues(adapter);
1534
1535         ixgbevf_configure(adapter);
1536
1537         ixgbevf_up_complete(adapter);
1538
1539         /* clear any pending interrupts, may auto mask */
1540         IXGBE_READ_REG(hw, IXGBE_VTEICR);
1541
1542         ixgbevf_irq_enable(adapter);
1543 }
1544
1545 /**
1546  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
1547  * @adapter: board private structure
1548  * @rx_ring: ring to free buffers from
1549  **/
1550 static void ixgbevf_clean_rx_ring(struct ixgbevf_adapter *adapter,
1551                                   struct ixgbevf_ring *rx_ring)
1552 {
1553         struct pci_dev *pdev = adapter->pdev;
1554         unsigned long size;
1555         unsigned int i;
1556
1557         if (!rx_ring->rx_buffer_info)
1558                 return;
1559
1560         /* Free all the Rx ring sk_buffs */
1561         for (i = 0; i < rx_ring->count; i++) {
1562                 struct ixgbevf_rx_buffer *rx_buffer_info;
1563
1564                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1565                 if (rx_buffer_info->dma) {
1566                         dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
1567                                          rx_ring->rx_buf_len,
1568                                          DMA_FROM_DEVICE);
1569                         rx_buffer_info->dma = 0;
1570                 }
1571                 if (rx_buffer_info->skb) {
1572                         struct sk_buff *skb = rx_buffer_info->skb;
1573                         rx_buffer_info->skb = NULL;
1574                         do {
1575                                 struct sk_buff *this = skb;
1576                                 skb = IXGBE_CB(skb)->prev;
1577                                 dev_kfree_skb(this);
1578                         } while (skb);
1579                 }
1580         }
1581
1582         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
1583         memset(rx_ring->rx_buffer_info, 0, size);
1584
1585         /* Zero out the descriptor ring */
1586         memset(rx_ring->desc, 0, rx_ring->size);
1587
1588         rx_ring->next_to_clean = 0;
1589         rx_ring->next_to_use = 0;
1590
1591         if (rx_ring->head)
1592                 writel(0, adapter->hw.hw_addr + rx_ring->head);
1593         if (rx_ring->tail)
1594                 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1595 }
1596
1597 /**
1598  * ixgbevf_clean_tx_ring - Free Tx Buffers
1599  * @adapter: board private structure
1600  * @tx_ring: ring to be cleaned
1601  **/
1602 static void ixgbevf_clean_tx_ring(struct ixgbevf_adapter *adapter,
1603                                   struct ixgbevf_ring *tx_ring)
1604 {
1605         struct ixgbevf_tx_buffer *tx_buffer_info;
1606         unsigned long size;
1607         unsigned int i;
1608
1609         if (!tx_ring->tx_buffer_info)
1610                 return;
1611
1612         /* Free all the Tx ring sk_buffs */
1613         for (i = 0; i < tx_ring->count; i++) {
1614                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
1615                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
1616         }
1617
1618         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
1619         memset(tx_ring->tx_buffer_info, 0, size);
1620
1621         memset(tx_ring->desc, 0, tx_ring->size);
1622
1623         tx_ring->next_to_use = 0;
1624         tx_ring->next_to_clean = 0;
1625
1626         if (tx_ring->head)
1627                 writel(0, adapter->hw.hw_addr + tx_ring->head);
1628         if (tx_ring->tail)
1629                 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1630 }
1631
1632 /**
1633  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
1634  * @adapter: board private structure
1635  **/
1636 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
1637 {
1638         int i;
1639
1640         for (i = 0; i < adapter->num_rx_queues; i++)
1641                 ixgbevf_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1642 }
1643
1644 /**
1645  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
1646  * @adapter: board private structure
1647  **/
1648 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
1649 {
1650         int i;
1651
1652         for (i = 0; i < adapter->num_tx_queues; i++)
1653                 ixgbevf_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1654 }
1655
1656 void ixgbevf_down(struct ixgbevf_adapter *adapter)
1657 {
1658         struct net_device *netdev = adapter->netdev;
1659         struct ixgbe_hw *hw = &adapter->hw;
1660         u32 txdctl;
1661         int i, j;
1662
1663         /* signal that we are down to the interrupt handler */
1664         set_bit(__IXGBEVF_DOWN, &adapter->state);
1665         /* disable receives */
1666
1667         netif_tx_disable(netdev);
1668
1669         msleep(10);
1670
1671         netif_tx_stop_all_queues(netdev);
1672
1673         ixgbevf_irq_disable(adapter);
1674
1675         ixgbevf_napi_disable_all(adapter);
1676
1677         del_timer_sync(&adapter->watchdog_timer);
1678         /* can't call flush scheduled work here because it can deadlock
1679          * if linkwatch_event tries to acquire the rtnl_lock which we are
1680          * holding */
1681         while (adapter->flags & IXGBE_FLAG_IN_WATCHDOG_TASK)
1682                 msleep(1);
1683
1684         /* disable transmits in the hardware now that interrupts are off */
1685         for (i = 0; i < adapter->num_tx_queues; i++) {
1686                 j = adapter->tx_ring[i].reg_idx;
1687                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1688                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j),
1689                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
1690         }
1691
1692         netif_carrier_off(netdev);
1693
1694         if (!pci_channel_offline(adapter->pdev))
1695                 ixgbevf_reset(adapter);
1696
1697         ixgbevf_clean_all_tx_rings(adapter);
1698         ixgbevf_clean_all_rx_rings(adapter);
1699 }
1700
1701 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
1702 {
1703         WARN_ON(in_interrupt());
1704
1705         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
1706                 msleep(1);
1707
1708         ixgbevf_down(adapter);
1709         ixgbevf_up(adapter);
1710
1711         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
1712 }
1713
1714 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
1715 {
1716         struct ixgbe_hw *hw = &adapter->hw;
1717         struct net_device *netdev = adapter->netdev;
1718
1719         if (hw->mac.ops.reset_hw(hw))
1720                 hw_dbg(hw, "PF still resetting\n");
1721         else
1722                 hw->mac.ops.init_hw(hw);
1723
1724         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1725                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1726                        netdev->addr_len);
1727                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1728                        netdev->addr_len);
1729         }
1730 }
1731
1732 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
1733                                         int vectors)
1734 {
1735         int err = 0;
1736         int vector_threshold;
1737
1738         /* We'll want at least 2 (vector_threshold):
1739          * 1) TxQ[0] + RxQ[0] handler
1740          * 2) Other (Link Status Change, etc.)
1741          */
1742         vector_threshold = MIN_MSIX_COUNT;
1743
1744         /* The more we get, the more we will assign to Tx/Rx Cleanup
1745          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
1746          * Right now, we simply care about how many we'll get; we'll
1747          * set them up later while requesting irq's.
1748          */
1749         while (vectors >= vector_threshold) {
1750                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
1751                                       vectors);
1752                 if (!err || err < 0) /* Success or a nasty failure. */
1753                         break;
1754                 else /* err == number of vectors we should try again with */
1755                         vectors = err;
1756         }
1757
1758         if (vectors < vector_threshold)
1759                 err = -ENOMEM;
1760
1761         if (err) {
1762                 dev_err(&adapter->pdev->dev,
1763                         "Unable to allocate MSI-X interrupts\n");
1764                 kfree(adapter->msix_entries);
1765                 adapter->msix_entries = NULL;
1766         } else {
1767                 /*
1768                  * Adjust for only the vectors we'll use, which is minimum
1769                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
1770                  * vectors we were allocated.
1771                  */
1772                 adapter->num_msix_vectors = vectors;
1773         }
1774
1775         return err;
1776 }
1777
1778 /**
1779  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
1780  * @adapter: board private structure to initialize
1781  *
1782  * This is the top level queue allocation routine.  The order here is very
1783  * important, starting with the "most" number of features turned on at once,
1784  * and ending with the smallest set of features.  This way large combinations
1785  * can be allocated if they're turned on, and smaller combinations are the
1786  * fallthrough conditions.
1787  *
1788  **/
1789 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
1790 {
1791         /* Start with base case */
1792         adapter->num_rx_queues = 1;
1793         adapter->num_tx_queues = 1;
1794 }
1795
1796 /**
1797  * ixgbevf_alloc_queues - Allocate memory for all rings
1798  * @adapter: board private structure to initialize
1799  *
1800  * We allocate one ring per queue at run-time since we don't know the
1801  * number of queues at compile-time.  The polling_netdev array is
1802  * intended for Multiqueue, but should work fine with a single queue.
1803  **/
1804 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
1805 {
1806         int i;
1807
1808         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1809                                    sizeof(struct ixgbevf_ring), GFP_KERNEL);
1810         if (!adapter->tx_ring)
1811                 goto err_tx_ring_allocation;
1812
1813         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1814                                    sizeof(struct ixgbevf_ring), GFP_KERNEL);
1815         if (!adapter->rx_ring)
1816                 goto err_rx_ring_allocation;
1817
1818         for (i = 0; i < adapter->num_tx_queues; i++) {
1819                 adapter->tx_ring[i].count = adapter->tx_ring_count;
1820                 adapter->tx_ring[i].queue_index = i;
1821                 /* reg_idx may be remapped later by DCB config */
1822                 adapter->tx_ring[i].reg_idx = i;
1823                 adapter->tx_ring[i].dev = &adapter->pdev->dev;
1824                 adapter->tx_ring[i].netdev = adapter->netdev;
1825         }
1826
1827         for (i = 0; i < adapter->num_rx_queues; i++) {
1828                 adapter->rx_ring[i].count = adapter->rx_ring_count;
1829                 adapter->rx_ring[i].queue_index = i;
1830                 adapter->rx_ring[i].reg_idx = i;
1831                 adapter->rx_ring[i].dev = &adapter->pdev->dev;
1832                 adapter->rx_ring[i].netdev = adapter->netdev;
1833         }
1834
1835         return 0;
1836
1837 err_rx_ring_allocation:
1838         kfree(adapter->tx_ring);
1839 err_tx_ring_allocation:
1840         return -ENOMEM;
1841 }
1842
1843 /**
1844  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
1845  * @adapter: board private structure to initialize
1846  *
1847  * Attempt to configure the interrupts using the best available
1848  * capabilities of the hardware and the kernel.
1849  **/
1850 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
1851 {
1852         struct net_device *netdev = adapter->netdev;
1853         int err = 0;
1854         int vector, v_budget;
1855
1856         /*
1857          * It's easy to be greedy for MSI-X vectors, but it really
1858          * doesn't do us much good if we have a lot more vectors
1859          * than CPU's.  So let's be conservative and only ask for
1860          * (roughly) the same number of vectors as there are CPU's.
1861          * The default is to use pairs of vectors.
1862          */
1863         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
1864         v_budget = min_t(int, v_budget, num_online_cpus());
1865         v_budget += NON_Q_VECTORS;
1866
1867         /* A failure in MSI-X entry allocation isn't fatal, but it does
1868          * mean we disable MSI-X capabilities of the adapter. */
1869         adapter->msix_entries = kcalloc(v_budget,
1870                                         sizeof(struct msix_entry), GFP_KERNEL);
1871         if (!adapter->msix_entries) {
1872                 err = -ENOMEM;
1873                 goto out;
1874         }
1875
1876         for (vector = 0; vector < v_budget; vector++)
1877                 adapter->msix_entries[vector].entry = vector;
1878
1879         err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
1880         if (err)
1881                 goto out;
1882
1883         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
1884         if (err)
1885                 goto out;
1886
1887         err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
1888
1889 out:
1890         return err;
1891 }
1892
1893 /**
1894  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
1895  * @adapter: board private structure to initialize
1896  *
1897  * We allocate one q_vector per queue interrupt.  If allocation fails we
1898  * return -ENOMEM.
1899  **/
1900 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
1901 {
1902         int q_idx, num_q_vectors;
1903         struct ixgbevf_q_vector *q_vector;
1904
1905         num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1906
1907         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1908                 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
1909                 if (!q_vector)
1910                         goto err_out;
1911                 q_vector->adapter = adapter;
1912                 q_vector->v_idx = q_idx;
1913                 netif_napi_add(adapter->netdev, &q_vector->napi,
1914                                ixgbevf_poll, 64);
1915                 adapter->q_vector[q_idx] = q_vector;
1916         }
1917
1918         return 0;
1919
1920 err_out:
1921         while (q_idx) {
1922                 q_idx--;
1923                 q_vector = adapter->q_vector[q_idx];
1924                 netif_napi_del(&q_vector->napi);
1925                 kfree(q_vector);
1926                 adapter->q_vector[q_idx] = NULL;
1927         }
1928         return -ENOMEM;
1929 }
1930
1931 /**
1932  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
1933  * @adapter: board private structure to initialize
1934  *
1935  * This function frees the memory allocated to the q_vectors.  In addition if
1936  * NAPI is enabled it will delete any references to the NAPI struct prior
1937  * to freeing the q_vector.
1938  **/
1939 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
1940 {
1941         int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1942
1943         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1944                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
1945
1946                 adapter->q_vector[q_idx] = NULL;
1947                 netif_napi_del(&q_vector->napi);
1948                 kfree(q_vector);
1949         }
1950 }
1951
1952 /**
1953  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
1954  * @adapter: board private structure
1955  *
1956  **/
1957 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
1958 {
1959         pci_disable_msix(adapter->pdev);
1960         kfree(adapter->msix_entries);
1961         adapter->msix_entries = NULL;
1962 }
1963
1964 /**
1965  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
1966  * @adapter: board private structure to initialize
1967  *
1968  **/
1969 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
1970 {
1971         int err;
1972
1973         /* Number of supported queues */
1974         ixgbevf_set_num_queues(adapter);
1975
1976         err = ixgbevf_set_interrupt_capability(adapter);
1977         if (err) {
1978                 hw_dbg(&adapter->hw,
1979                        "Unable to setup interrupt capabilities\n");
1980                 goto err_set_interrupt;
1981         }
1982
1983         err = ixgbevf_alloc_q_vectors(adapter);
1984         if (err) {
1985                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
1986                        "vectors\n");
1987                 goto err_alloc_q_vectors;
1988         }
1989
1990         err = ixgbevf_alloc_queues(adapter);
1991         if (err) {
1992                 pr_err("Unable to allocate memory for queues\n");
1993                 goto err_alloc_queues;
1994         }
1995
1996         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
1997                "Tx Queue count = %u\n",
1998                (adapter->num_rx_queues > 1) ? "Enabled" :
1999                "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2000
2001         set_bit(__IXGBEVF_DOWN, &adapter->state);
2002
2003         return 0;
2004 err_alloc_queues:
2005         ixgbevf_free_q_vectors(adapter);
2006 err_alloc_q_vectors:
2007         ixgbevf_reset_interrupt_capability(adapter);
2008 err_set_interrupt:
2009         return err;
2010 }
2011
2012 /**
2013  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2014  * @adapter: board private structure to clear interrupt scheme on
2015  *
2016  * We go through and clear interrupt specific resources and reset the structure
2017  * to pre-load conditions
2018  **/
2019 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2020 {
2021         adapter->num_tx_queues = 0;
2022         adapter->num_rx_queues = 0;
2023
2024         ixgbevf_free_q_vectors(adapter);
2025         ixgbevf_reset_interrupt_capability(adapter);
2026 }
2027
2028 /**
2029  * ixgbevf_sw_init - Initialize general software structures
2030  * (struct ixgbevf_adapter)
2031  * @adapter: board private structure to initialize
2032  *
2033  * ixgbevf_sw_init initializes the Adapter private data structure.
2034  * Fields are initialized based on PCI device information and
2035  * OS network device settings (MTU size).
2036  **/
2037 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2038 {
2039         struct ixgbe_hw *hw = &adapter->hw;
2040         struct pci_dev *pdev = adapter->pdev;
2041         int err;
2042
2043         /* PCI config space info */
2044
2045         hw->vendor_id = pdev->vendor;
2046         hw->device_id = pdev->device;
2047         hw->revision_id = pdev->revision;
2048         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2049         hw->subsystem_device_id = pdev->subsystem_device;
2050
2051         hw->mbx.ops.init_params(hw);
2052
2053         /* assume legacy case in which PF would only give VF 2 queues */
2054         hw->mac.max_tx_queues = 2;
2055         hw->mac.max_rx_queues = 2;
2056
2057         err = hw->mac.ops.reset_hw(hw);
2058         if (err) {
2059                 dev_info(&pdev->dev,
2060                          "PF still in reset state, assigning new address\n");
2061                 eth_hw_addr_random(adapter->netdev);
2062                 memcpy(adapter->hw.mac.addr, adapter->netdev->dev_addr,
2063                         adapter->netdev->addr_len);
2064         } else {
2065                 err = hw->mac.ops.init_hw(hw);
2066                 if (err) {
2067                         pr_err("init_shared_code failed: %d\n", err);
2068                         goto out;
2069                 }
2070                 memcpy(adapter->netdev->dev_addr, adapter->hw.mac.addr,
2071                        adapter->netdev->addr_len);
2072         }
2073
2074         /* lock to protect mailbox accesses */
2075         spin_lock_init(&adapter->mbx_lock);
2076
2077         /* Enable dynamic interrupt throttling rates */
2078         adapter->rx_itr_setting = 1;
2079         adapter->tx_itr_setting = 1;
2080
2081         /* set default ring sizes */
2082         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2083         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2084
2085         set_bit(__IXGBEVF_DOWN, &adapter->state);
2086         return 0;
2087
2088 out:
2089         return err;
2090 }
2091
2092 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
2093         {                                                       \
2094                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
2095                 if (current_counter < last_counter)             \
2096                         counter += 0x100000000LL;               \
2097                 last_counter = current_counter;                 \
2098                 counter &= 0xFFFFFFFF00000000LL;                \
2099                 counter |= current_counter;                     \
2100         }
2101
2102 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2103         {                                                                \
2104                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
2105                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
2106                 u64 current_counter = (current_counter_msb << 32) |      \
2107                         current_counter_lsb;                             \
2108                 if (current_counter < last_counter)                      \
2109                         counter += 0x1000000000LL;                       \
2110                 last_counter = current_counter;                          \
2111                 counter &= 0xFFFFFFF000000000LL;                         \
2112                 counter |= current_counter;                              \
2113         }
2114 /**
2115  * ixgbevf_update_stats - Update the board statistics counters.
2116  * @adapter: board private structure
2117  **/
2118 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2119 {
2120         struct ixgbe_hw *hw = &adapter->hw;
2121         int i;
2122
2123         if (!adapter->link_up)
2124                 return;
2125
2126         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2127                                 adapter->stats.vfgprc);
2128         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2129                                 adapter->stats.vfgptc);
2130         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2131                                 adapter->stats.last_vfgorc,
2132                                 adapter->stats.vfgorc);
2133         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2134                                 adapter->stats.last_vfgotc,
2135                                 adapter->stats.vfgotc);
2136         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2137                                 adapter->stats.vfmprc);
2138
2139         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2140                 adapter->hw_csum_rx_error +=
2141                         adapter->rx_ring[i].hw_csum_rx_error;
2142                 adapter->hw_csum_rx_good +=
2143                         adapter->rx_ring[i].hw_csum_rx_good;
2144                 adapter->rx_ring[i].hw_csum_rx_error = 0;
2145                 adapter->rx_ring[i].hw_csum_rx_good = 0;
2146         }
2147 }
2148
2149 /**
2150  * ixgbevf_watchdog - Timer Call-back
2151  * @data: pointer to adapter cast into an unsigned long
2152  **/
2153 static void ixgbevf_watchdog(unsigned long data)
2154 {
2155         struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2156         struct ixgbe_hw *hw = &adapter->hw;
2157         u32 eics = 0;
2158         int i;
2159
2160         /*
2161          * Do the watchdog outside of interrupt context due to the lovely
2162          * delays that some of the newer hardware requires
2163          */
2164
2165         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
2166                 goto watchdog_short_circuit;
2167
2168         /* get one bit for every active tx/rx interrupt vector */
2169         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2170                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2171                 if (qv->rx.ring || qv->tx.ring)
2172                         eics |= 1 << i;
2173         }
2174
2175         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2176
2177 watchdog_short_circuit:
2178         schedule_work(&adapter->watchdog_task);
2179 }
2180
2181 /**
2182  * ixgbevf_tx_timeout - Respond to a Tx Hang
2183  * @netdev: network interface device structure
2184  **/
2185 static void ixgbevf_tx_timeout(struct net_device *netdev)
2186 {
2187         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2188
2189         /* Do the reset outside of interrupt context */
2190         schedule_work(&adapter->reset_task);
2191 }
2192
2193 static void ixgbevf_reset_task(struct work_struct *work)
2194 {
2195         struct ixgbevf_adapter *adapter;
2196         adapter = container_of(work, struct ixgbevf_adapter, reset_task);
2197
2198         /* If we're already down or resetting, just bail */
2199         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2200             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2201                 return;
2202
2203         adapter->tx_timeout_count++;
2204
2205         ixgbevf_reinit_locked(adapter);
2206 }
2207
2208 /**
2209  * ixgbevf_watchdog_task - worker thread to bring link up
2210  * @work: pointer to work_struct containing our data
2211  **/
2212 static void ixgbevf_watchdog_task(struct work_struct *work)
2213 {
2214         struct ixgbevf_adapter *adapter = container_of(work,
2215                                                        struct ixgbevf_adapter,
2216                                                        watchdog_task);
2217         struct net_device *netdev = adapter->netdev;
2218         struct ixgbe_hw *hw = &adapter->hw;
2219         u32 link_speed = adapter->link_speed;
2220         bool link_up = adapter->link_up;
2221         s32 need_reset;
2222
2223         adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
2224
2225         /*
2226          * Always check the link on the watchdog because we have
2227          * no LSC interrupt
2228          */
2229         spin_lock_bh(&adapter->mbx_lock);
2230
2231         need_reset = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2232
2233         spin_unlock_bh(&adapter->mbx_lock);
2234
2235         if (need_reset) {
2236                 adapter->link_up = link_up;
2237                 adapter->link_speed = link_speed;
2238                 netif_carrier_off(netdev);
2239                 netif_tx_stop_all_queues(netdev);
2240                 schedule_work(&adapter->reset_task);
2241                 goto pf_has_reset;
2242         }
2243         adapter->link_up = link_up;
2244         adapter->link_speed = link_speed;
2245
2246         if (link_up) {
2247                 if (!netif_carrier_ok(netdev)) {
2248                         char *link_speed_string;
2249                         switch (link_speed) {
2250                         case IXGBE_LINK_SPEED_10GB_FULL:
2251                                 link_speed_string = "10 Gbps";
2252                                 break;
2253                         case IXGBE_LINK_SPEED_1GB_FULL:
2254                                 link_speed_string = "1 Gbps";
2255                                 break;
2256                         case IXGBE_LINK_SPEED_100_FULL:
2257                                 link_speed_string = "100 Mbps";
2258                                 break;
2259                         default:
2260                                 link_speed_string = "unknown speed";
2261                                 break;
2262                         }
2263                         dev_info(&adapter->pdev->dev,
2264                                 "NIC Link is Up, %s\n", link_speed_string);
2265                         netif_carrier_on(netdev);
2266                         netif_tx_wake_all_queues(netdev);
2267                 }
2268         } else {
2269                 adapter->link_up = false;
2270                 adapter->link_speed = 0;
2271                 if (netif_carrier_ok(netdev)) {
2272                         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2273                         netif_carrier_off(netdev);
2274                         netif_tx_stop_all_queues(netdev);
2275                 }
2276         }
2277
2278         ixgbevf_update_stats(adapter);
2279
2280 pf_has_reset:
2281         /* Reset the timer */
2282         if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
2283                 mod_timer(&adapter->watchdog_timer,
2284                           round_jiffies(jiffies + (2 * HZ)));
2285
2286         adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
2287 }
2288
2289 /**
2290  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2291  * @adapter: board private structure
2292  * @tx_ring: Tx descriptor ring for a specific queue
2293  *
2294  * Free all transmit software resources
2295  **/
2296 void ixgbevf_free_tx_resources(struct ixgbevf_adapter *adapter,
2297                                struct ixgbevf_ring *tx_ring)
2298 {
2299         struct pci_dev *pdev = adapter->pdev;
2300
2301         ixgbevf_clean_tx_ring(adapter, tx_ring);
2302
2303         vfree(tx_ring->tx_buffer_info);
2304         tx_ring->tx_buffer_info = NULL;
2305
2306         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2307                           tx_ring->dma);
2308
2309         tx_ring->desc = NULL;
2310 }
2311
2312 /**
2313  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2314  * @adapter: board private structure
2315  *
2316  * Free all transmit software resources
2317  **/
2318 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2319 {
2320         int i;
2321
2322         for (i = 0; i < adapter->num_tx_queues; i++)
2323                 if (adapter->tx_ring[i].desc)
2324                         ixgbevf_free_tx_resources(adapter,
2325                                                   &adapter->tx_ring[i]);
2326
2327 }
2328
2329 /**
2330  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2331  * @adapter: board private structure
2332  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
2333  *
2334  * Return 0 on success, negative on failure
2335  **/
2336 int ixgbevf_setup_tx_resources(struct ixgbevf_adapter *adapter,
2337                                struct ixgbevf_ring *tx_ring)
2338 {
2339         struct pci_dev *pdev = adapter->pdev;
2340         int size;
2341
2342         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2343         tx_ring->tx_buffer_info = vzalloc(size);
2344         if (!tx_ring->tx_buffer_info)
2345                 goto err;
2346
2347         /* round up to nearest 4K */
2348         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2349         tx_ring->size = ALIGN(tx_ring->size, 4096);
2350
2351         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
2352                                            &tx_ring->dma, GFP_KERNEL);
2353         if (!tx_ring->desc)
2354                 goto err;
2355
2356         tx_ring->next_to_use = 0;
2357         tx_ring->next_to_clean = 0;
2358         return 0;
2359
2360 err:
2361         vfree(tx_ring->tx_buffer_info);
2362         tx_ring->tx_buffer_info = NULL;
2363         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2364                "descriptor ring\n");
2365         return -ENOMEM;
2366 }
2367
2368 /**
2369  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2370  * @adapter: board private structure
2371  *
2372  * If this function returns with an error, then it's possible one or
2373  * more of the rings is populated (while the rest are not).  It is the
2374  * callers duty to clean those orphaned rings.
2375  *
2376  * Return 0 on success, negative on failure
2377  **/
2378 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2379 {
2380         int i, err = 0;
2381
2382         for (i = 0; i < adapter->num_tx_queues; i++) {
2383                 err = ixgbevf_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2384                 if (!err)
2385                         continue;
2386                 hw_dbg(&adapter->hw,
2387                        "Allocation for Tx Queue %u failed\n", i);
2388                 break;
2389         }
2390
2391         return err;
2392 }
2393
2394 /**
2395  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2396  * @adapter: board private structure
2397  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
2398  *
2399  * Returns 0 on success, negative on failure
2400  **/
2401 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
2402                                struct ixgbevf_ring *rx_ring)
2403 {
2404         struct pci_dev *pdev = adapter->pdev;
2405         int size;
2406
2407         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2408         rx_ring->rx_buffer_info = vzalloc(size);
2409         if (!rx_ring->rx_buffer_info)
2410                 goto alloc_failed;
2411
2412         /* Round up to nearest 4K */
2413         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2414         rx_ring->size = ALIGN(rx_ring->size, 4096);
2415
2416         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
2417                                            &rx_ring->dma, GFP_KERNEL);
2418
2419         if (!rx_ring->desc) {
2420                 hw_dbg(&adapter->hw,
2421                        "Unable to allocate memory for "
2422                        "the receive descriptor ring\n");
2423                 vfree(rx_ring->rx_buffer_info);
2424                 rx_ring->rx_buffer_info = NULL;
2425                 goto alloc_failed;
2426         }
2427
2428         rx_ring->next_to_clean = 0;
2429         rx_ring->next_to_use = 0;
2430
2431         return 0;
2432 alloc_failed:
2433         return -ENOMEM;
2434 }
2435
2436 /**
2437  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
2438  * @adapter: board private structure
2439  *
2440  * If this function returns with an error, then it's possible one or
2441  * more of the rings is populated (while the rest are not).  It is the
2442  * callers duty to clean those orphaned rings.
2443  *
2444  * Return 0 on success, negative on failure
2445  **/
2446 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
2447 {
2448         int i, err = 0;
2449
2450         for (i = 0; i < adapter->num_rx_queues; i++) {
2451                 err = ixgbevf_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2452                 if (!err)
2453                         continue;
2454                 hw_dbg(&adapter->hw,
2455                        "Allocation for Rx Queue %u failed\n", i);
2456                 break;
2457         }
2458         return err;
2459 }
2460
2461 /**
2462  * ixgbevf_free_rx_resources - Free Rx Resources
2463  * @adapter: board private structure
2464  * @rx_ring: ring to clean the resources from
2465  *
2466  * Free all receive software resources
2467  **/
2468 void ixgbevf_free_rx_resources(struct ixgbevf_adapter *adapter,
2469                                struct ixgbevf_ring *rx_ring)
2470 {
2471         struct pci_dev *pdev = adapter->pdev;
2472
2473         ixgbevf_clean_rx_ring(adapter, rx_ring);
2474
2475         vfree(rx_ring->rx_buffer_info);
2476         rx_ring->rx_buffer_info = NULL;
2477
2478         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2479                           rx_ring->dma);
2480
2481         rx_ring->desc = NULL;
2482 }
2483
2484 /**
2485  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
2486  * @adapter: board private structure
2487  *
2488  * Free all receive software resources
2489  **/
2490 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
2491 {
2492         int i;
2493
2494         for (i = 0; i < adapter->num_rx_queues; i++)
2495                 if (adapter->rx_ring[i].desc)
2496                         ixgbevf_free_rx_resources(adapter,
2497                                                   &adapter->rx_ring[i]);
2498 }
2499
2500 static int ixgbevf_setup_queues(struct ixgbevf_adapter *adapter)
2501 {
2502         struct ixgbe_hw *hw = &adapter->hw;
2503         struct ixgbevf_ring *rx_ring;
2504         unsigned int def_q = 0;
2505         unsigned int num_tcs = 0;
2506         unsigned int num_rx_queues = 1;
2507         int err, i;
2508
2509         spin_lock_bh(&adapter->mbx_lock);
2510
2511         /* fetch queue configuration from the PF */
2512         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2513
2514         spin_unlock_bh(&adapter->mbx_lock);
2515
2516         if (err)
2517                 return err;
2518
2519         if (num_tcs > 1) {
2520                 /* update default Tx ring register index */
2521                 adapter->tx_ring[0].reg_idx = def_q;
2522
2523                 /* we need as many queues as traffic classes */
2524                 num_rx_queues = num_tcs;
2525         }
2526
2527         /* nothing to do if we have the correct number of queues */
2528         if (adapter->num_rx_queues == num_rx_queues)
2529                 return 0;
2530
2531         /* allocate new rings */
2532         rx_ring = kcalloc(num_rx_queues,
2533                           sizeof(struct ixgbevf_ring), GFP_KERNEL);
2534         if (!rx_ring)
2535                 return -ENOMEM;
2536
2537         /* setup ring fields */
2538         for (i = 0; i < num_rx_queues; i++) {
2539                 rx_ring[i].count = adapter->rx_ring_count;
2540                 rx_ring[i].queue_index = i;
2541                 rx_ring[i].reg_idx = i;
2542                 rx_ring[i].dev = &adapter->pdev->dev;
2543                 rx_ring[i].netdev = adapter->netdev;
2544         }
2545
2546         /* free the existing ring and queues */
2547         adapter->num_rx_queues = 0;
2548         kfree(adapter->rx_ring);
2549
2550         /* move new rings into position on the adapter struct */
2551         adapter->rx_ring = rx_ring;
2552         adapter->num_rx_queues = num_rx_queues;
2553
2554         return 0;
2555 }
2556
2557 /**
2558  * ixgbevf_open - Called when a network interface is made active
2559  * @netdev: network interface device structure
2560  *
2561  * Returns 0 on success, negative value on failure
2562  *
2563  * The open entry point is called when a network interface is made
2564  * active by the system (IFF_UP).  At this point all resources needed
2565  * for transmit and receive operations are allocated, the interrupt
2566  * handler is registered with the OS, the watchdog timer is started,
2567  * and the stack is notified that the interface is ready.
2568  **/
2569 static int ixgbevf_open(struct net_device *netdev)
2570 {
2571         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2572         struct ixgbe_hw *hw = &adapter->hw;
2573         int err;
2574
2575         /* disallow open during test */
2576         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
2577                 return -EBUSY;
2578
2579         if (hw->adapter_stopped) {
2580                 ixgbevf_reset(adapter);
2581                 /* if adapter is still stopped then PF isn't up and
2582                  * the vf can't start. */
2583                 if (hw->adapter_stopped) {
2584                         err = IXGBE_ERR_MBX;
2585                         pr_err("Unable to start - perhaps the PF Driver isn't "
2586                                "up yet\n");
2587                         goto err_setup_reset;
2588                 }
2589         }
2590
2591         ixgbevf_negotiate_api(adapter);
2592
2593         /* setup queue reg_idx and Rx queue count */
2594         err = ixgbevf_setup_queues(adapter);
2595         if (err)
2596                 goto err_setup_queues;
2597
2598         /* allocate transmit descriptors */
2599         err = ixgbevf_setup_all_tx_resources(adapter);
2600         if (err)
2601                 goto err_setup_tx;
2602
2603         /* allocate receive descriptors */
2604         err = ixgbevf_setup_all_rx_resources(adapter);
2605         if (err)
2606                 goto err_setup_rx;
2607
2608         ixgbevf_configure(adapter);
2609
2610         /*
2611          * Map the Tx/Rx rings to the vectors we were allotted.
2612          * if request_irq will be called in this function map_rings
2613          * must be called *before* up_complete
2614          */
2615         ixgbevf_map_rings_to_vectors(adapter);
2616
2617         ixgbevf_up_complete(adapter);
2618
2619         /* clear any pending interrupts, may auto mask */
2620         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2621         err = ixgbevf_request_irq(adapter);
2622         if (err)
2623                 goto err_req_irq;
2624
2625         ixgbevf_irq_enable(adapter);
2626
2627         return 0;
2628
2629 err_req_irq:
2630         ixgbevf_down(adapter);
2631         ixgbevf_free_irq(adapter);
2632 err_setup_rx:
2633         ixgbevf_free_all_rx_resources(adapter);
2634 err_setup_tx:
2635         ixgbevf_free_all_tx_resources(adapter);
2636 err_setup_queues:
2637         ixgbevf_reset(adapter);
2638
2639 err_setup_reset:
2640
2641         return err;
2642 }
2643
2644 /**
2645  * ixgbevf_close - Disables a network interface
2646  * @netdev: network interface device structure
2647  *
2648  * Returns 0, this is not allowed to fail
2649  *
2650  * The close entry point is called when an interface is de-activated
2651  * by the OS.  The hardware is still under the drivers control, but
2652  * needs to be disabled.  A global MAC reset is issued to stop the
2653  * hardware, and all transmit and receive resources are freed.
2654  **/
2655 static int ixgbevf_close(struct net_device *netdev)
2656 {
2657         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2658
2659         ixgbevf_down(adapter);
2660         ixgbevf_free_irq(adapter);
2661
2662         ixgbevf_free_all_tx_resources(adapter);
2663         ixgbevf_free_all_rx_resources(adapter);
2664
2665         return 0;
2666 }
2667
2668 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
2669                                 u32 vlan_macip_lens, u32 type_tucmd,
2670                                 u32 mss_l4len_idx)
2671 {
2672         struct ixgbe_adv_tx_context_desc *context_desc;
2673         u16 i = tx_ring->next_to_use;
2674
2675         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
2676
2677         i++;
2678         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2679
2680         /* set bits to identify this as an advanced context descriptor */
2681         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
2682
2683         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
2684         context_desc->seqnum_seed       = 0;
2685         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
2686         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
2687 }
2688
2689 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
2690                        struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2691 {
2692         u32 vlan_macip_lens, type_tucmd;
2693         u32 mss_l4len_idx, l4len;
2694
2695         if (!skb_is_gso(skb))
2696                 return 0;
2697
2698         if (skb_header_cloned(skb)) {
2699                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2700                 if (err)
2701                         return err;
2702         }
2703
2704         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2705         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
2706
2707         if (skb->protocol == htons(ETH_P_IP)) {
2708                 struct iphdr *iph = ip_hdr(skb);
2709                 iph->tot_len = 0;
2710                 iph->check = 0;
2711                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2712                                                          iph->daddr, 0,
2713                                                          IPPROTO_TCP,
2714                                                          0);
2715                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2716         } else if (skb_is_gso_v6(skb)) {
2717                 ipv6_hdr(skb)->payload_len = 0;
2718                 tcp_hdr(skb)->check =
2719                     ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2720                                      &ipv6_hdr(skb)->daddr,
2721                                      0, IPPROTO_TCP, 0);
2722         }
2723
2724         /* compute header lengths */
2725         l4len = tcp_hdrlen(skb);
2726         *hdr_len += l4len;
2727         *hdr_len = skb_transport_offset(skb) + l4len;
2728
2729         /* mss_l4len_id: use 1 as index for TSO */
2730         mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
2731         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
2732         mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
2733
2734         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
2735         vlan_macip_lens = skb_network_header_len(skb);
2736         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2737         vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2738
2739         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2740                             type_tucmd, mss_l4len_idx);
2741
2742         return 1;
2743 }
2744
2745 static bool ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
2746                             struct sk_buff *skb, u32 tx_flags)
2747 {
2748         u32 vlan_macip_lens = 0;
2749         u32 mss_l4len_idx = 0;
2750         u32 type_tucmd = 0;
2751
2752         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2753                 u8 l4_hdr = 0;
2754                 switch (skb->protocol) {
2755                 case __constant_htons(ETH_P_IP):
2756                         vlan_macip_lens |= skb_network_header_len(skb);
2757                         type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2758                         l4_hdr = ip_hdr(skb)->protocol;
2759                         break;
2760                 case __constant_htons(ETH_P_IPV6):
2761                         vlan_macip_lens |= skb_network_header_len(skb);
2762                         l4_hdr = ipv6_hdr(skb)->nexthdr;
2763                         break;
2764                 default:
2765                         if (unlikely(net_ratelimit())) {
2766                                 dev_warn(tx_ring->dev,
2767                                  "partial checksum but proto=%x!\n",
2768                                  skb->protocol);
2769                         }
2770                         break;
2771                 }
2772
2773                 switch (l4_hdr) {
2774                 case IPPROTO_TCP:
2775                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
2776                         mss_l4len_idx = tcp_hdrlen(skb) <<
2777                                         IXGBE_ADVTXD_L4LEN_SHIFT;
2778                         break;
2779                 case IPPROTO_SCTP:
2780                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
2781                         mss_l4len_idx = sizeof(struct sctphdr) <<
2782                                         IXGBE_ADVTXD_L4LEN_SHIFT;
2783                         break;
2784                 case IPPROTO_UDP:
2785                         mss_l4len_idx = sizeof(struct udphdr) <<
2786                                         IXGBE_ADVTXD_L4LEN_SHIFT;
2787                         break;
2788                 default:
2789                         if (unlikely(net_ratelimit())) {
2790                                 dev_warn(tx_ring->dev,
2791                                  "partial checksum but l4 proto=%x!\n",
2792                                  l4_hdr);
2793                         }
2794                         break;
2795                 }
2796         }
2797
2798         /* vlan_macip_lens: MACLEN, VLAN tag */
2799         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2800         vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2801
2802         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2803                             type_tucmd, mss_l4len_idx);
2804
2805         return (skb->ip_summed == CHECKSUM_PARTIAL);
2806 }
2807
2808 static int ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
2809                           struct sk_buff *skb, u32 tx_flags,
2810                           unsigned int first)
2811 {
2812         struct ixgbevf_tx_buffer *tx_buffer_info;
2813         unsigned int len;
2814         unsigned int total = skb->len;
2815         unsigned int offset = 0, size;
2816         int count = 0;
2817         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
2818         unsigned int f;
2819         int i;
2820
2821         i = tx_ring->next_to_use;
2822
2823         len = min(skb_headlen(skb), total);
2824         while (len) {
2825                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2826                 size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2827
2828                 tx_buffer_info->length = size;
2829                 tx_buffer_info->mapped_as_page = false;
2830                 tx_buffer_info->dma = dma_map_single(tx_ring->dev,
2831                                                      skb->data + offset,
2832                                                      size, DMA_TO_DEVICE);
2833                 if (dma_mapping_error(tx_ring->dev, tx_buffer_info->dma))
2834                         goto dma_error;
2835                 tx_buffer_info->next_to_watch = i;
2836
2837                 len -= size;
2838                 total -= size;
2839                 offset += size;
2840                 count++;
2841                 i++;
2842                 if (i == tx_ring->count)
2843                         i = 0;
2844         }
2845
2846         for (f = 0; f < nr_frags; f++) {
2847                 const struct skb_frag_struct *frag;
2848
2849                 frag = &skb_shinfo(skb)->frags[f];
2850                 len = min((unsigned int)skb_frag_size(frag), total);
2851                 offset = 0;
2852
2853                 while (len) {
2854                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
2855                         size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2856
2857                         tx_buffer_info->length = size;
2858                         tx_buffer_info->dma =
2859                                 skb_frag_dma_map(tx_ring->dev, frag,
2860                                                  offset, size, DMA_TO_DEVICE);
2861                         if (dma_mapping_error(tx_ring->dev,
2862                                               tx_buffer_info->dma))
2863                                 goto dma_error;
2864                         tx_buffer_info->mapped_as_page = true;
2865                         tx_buffer_info->next_to_watch = i;
2866
2867                         len -= size;
2868                         total -= size;
2869                         offset += size;
2870                         count++;
2871                         i++;
2872                         if (i == tx_ring->count)
2873                                 i = 0;
2874                 }
2875                 if (total == 0)
2876                         break;
2877         }
2878
2879         if (i == 0)
2880                 i = tx_ring->count - 1;
2881         else
2882                 i = i - 1;
2883         tx_ring->tx_buffer_info[i].skb = skb;
2884         tx_ring->tx_buffer_info[first].next_to_watch = i;
2885         tx_ring->tx_buffer_info[first].time_stamp = jiffies;
2886
2887         return count;
2888
2889 dma_error:
2890         dev_err(tx_ring->dev, "TX DMA map failed\n");
2891
2892         /* clear timestamp and dma mappings for failed tx_buffer_info map */
2893         tx_buffer_info->dma = 0;
2894         tx_buffer_info->next_to_watch = 0;
2895         count--;
2896
2897         /* clear timestamp and dma mappings for remaining portion of packet */
2898         while (count >= 0) {
2899                 count--;
2900                 i--;
2901                 if (i < 0)
2902                         i += tx_ring->count;
2903                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2904                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
2905         }
2906
2907         return count;
2908 }
2909
2910 static void ixgbevf_tx_queue(struct ixgbevf_ring *tx_ring, int tx_flags,
2911                              int count, u32 paylen, u8 hdr_len)
2912 {
2913         union ixgbe_adv_tx_desc *tx_desc = NULL;
2914         struct ixgbevf_tx_buffer *tx_buffer_info;
2915         u32 olinfo_status = 0, cmd_type_len = 0;
2916         unsigned int i;
2917
2918         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
2919
2920         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
2921
2922         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
2923
2924         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2925                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
2926
2927         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
2928                 olinfo_status |= IXGBE_ADVTXD_POPTS_TXSM;
2929
2930         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
2931                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
2932
2933                 /* use index 1 context for tso */
2934                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
2935                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
2936                         olinfo_status |= IXGBE_ADVTXD_POPTS_IXSM;
2937         }
2938
2939         /*
2940          * Check Context must be set if Tx switch is enabled, which it
2941          * always is for case where virtual functions are running
2942          */
2943         olinfo_status |= IXGBE_ADVTXD_CC;
2944
2945         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
2946
2947         i = tx_ring->next_to_use;
2948         while (count--) {
2949                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2950                 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
2951                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
2952                 tx_desc->read.cmd_type_len =
2953                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
2954                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2955                 i++;
2956                 if (i == tx_ring->count)
2957                         i = 0;
2958         }
2959
2960         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
2961
2962         tx_ring->next_to_use = i;
2963 }
2964
2965 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
2966 {
2967         struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
2968
2969         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
2970         /* Herbert's original patch had:
2971          *  smp_mb__after_netif_stop_queue();
2972          * but since that doesn't exist yet, just open code it. */
2973         smp_mb();
2974
2975         /* We need to check again in a case another CPU has just
2976          * made room available. */
2977         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
2978                 return -EBUSY;
2979
2980         /* A reprieve! - use start_queue because it doesn't call schedule */
2981         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
2982         ++adapter->restart_queue;
2983         return 0;
2984 }
2985
2986 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
2987 {
2988         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
2989                 return 0;
2990         return __ixgbevf_maybe_stop_tx(tx_ring, size);
2991 }
2992
2993 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2994 {
2995         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2996         struct ixgbevf_ring *tx_ring;
2997         unsigned int first;
2998         unsigned int tx_flags = 0;
2999         u8 hdr_len = 0;
3000         int r_idx = 0, tso;
3001         u16 count = TXD_USE_COUNT(skb_headlen(skb));
3002 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3003         unsigned short f;
3004 #endif
3005         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3006         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3007                 dev_kfree_skb(skb);
3008                 return NETDEV_TX_OK;
3009         }
3010
3011         tx_ring = &adapter->tx_ring[r_idx];
3012
3013         /*
3014          * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3015          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3016          *       + 2 desc gap to keep tail from touching head,
3017          *       + 1 desc for context descriptor,
3018          * otherwise try next time
3019          */
3020 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3021         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3022                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3023 #else
3024         count += skb_shinfo(skb)->nr_frags;
3025 #endif
3026         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3027                 adapter->tx_busy++;
3028                 return NETDEV_TX_BUSY;
3029         }
3030
3031         if (vlan_tx_tag_present(skb)) {
3032                 tx_flags |= vlan_tx_tag_get(skb);
3033                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3034                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3035         }
3036
3037         first = tx_ring->next_to_use;
3038
3039         if (skb->protocol == htons(ETH_P_IP))
3040                 tx_flags |= IXGBE_TX_FLAGS_IPV4;
3041         tso = ixgbevf_tso(tx_ring, skb, tx_flags, &hdr_len);
3042         if (tso < 0) {
3043                 dev_kfree_skb_any(skb);
3044                 return NETDEV_TX_OK;
3045         }
3046
3047         if (tso)
3048                 tx_flags |= IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_CSUM;
3049         else if (ixgbevf_tx_csum(tx_ring, skb, tx_flags))
3050                 tx_flags |= IXGBE_TX_FLAGS_CSUM;
3051
3052         ixgbevf_tx_queue(tx_ring, tx_flags,
3053                          ixgbevf_tx_map(tx_ring, skb, tx_flags, first),
3054                          skb->len, hdr_len);
3055         /*
3056          * Force memory writes to complete before letting h/w
3057          * know there are new descriptors to fetch.  (Only
3058          * applicable for weak-ordered memory model archs,
3059          * such as IA-64).
3060          */
3061         wmb();
3062
3063         writel(tx_ring->next_to_use, adapter->hw.hw_addr + tx_ring->tail);
3064
3065         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3066
3067         return NETDEV_TX_OK;
3068 }
3069
3070 /**
3071  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3072  * @netdev: network interface device structure
3073  * @p: pointer to an address structure
3074  *
3075  * Returns 0 on success, negative on failure
3076  **/
3077 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3078 {
3079         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3080         struct ixgbe_hw *hw = &adapter->hw;
3081         struct sockaddr *addr = p;
3082
3083         if (!is_valid_ether_addr(addr->sa_data))
3084                 return -EADDRNOTAVAIL;
3085
3086         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3087         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3088
3089         spin_lock_bh(&adapter->mbx_lock);
3090
3091         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3092
3093         spin_unlock_bh(&adapter->mbx_lock);
3094
3095         return 0;
3096 }
3097
3098 /**
3099  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3100  * @netdev: network interface device structure
3101  * @new_mtu: new value for maximum frame size
3102  *
3103  * Returns 0 on success, negative on failure
3104  **/
3105 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3106 {
3107         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3108         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3109         int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3110
3111         switch (adapter->hw.api_version) {
3112         case ixgbe_mbox_api_11:
3113                 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3114                 break;
3115         default:
3116                 if (adapter->hw.mac.type == ixgbe_mac_X540_vf)
3117                         max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3118                 break;
3119         }
3120
3121         /* MTU < 68 is an error and causes problems on some kernels */
3122         if ((new_mtu < 68) || (max_frame > max_possible_frame))
3123                 return -EINVAL;
3124
3125         hw_dbg(&adapter->hw, "changing MTU from %d to %d\n",
3126                netdev->mtu, new_mtu);
3127         /* must set new MTU before calling down or up */
3128         netdev->mtu = new_mtu;
3129
3130         if (netif_running(netdev))
3131                 ixgbevf_reinit_locked(adapter);
3132
3133         return 0;
3134 }
3135
3136 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3137 {
3138         struct net_device *netdev = pci_get_drvdata(pdev);
3139         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3140 #ifdef CONFIG_PM
3141         int retval = 0;
3142 #endif
3143
3144         netif_device_detach(netdev);
3145
3146         if (netif_running(netdev)) {
3147                 rtnl_lock();
3148                 ixgbevf_down(adapter);
3149                 ixgbevf_free_irq(adapter);
3150                 ixgbevf_free_all_tx_resources(adapter);
3151                 ixgbevf_free_all_rx_resources(adapter);
3152                 rtnl_unlock();
3153         }
3154
3155         ixgbevf_clear_interrupt_scheme(adapter);
3156
3157 #ifdef CONFIG_PM
3158         retval = pci_save_state(pdev);
3159         if (retval)
3160                 return retval;
3161
3162 #endif
3163         pci_disable_device(pdev);
3164
3165         return 0;
3166 }
3167
3168 #ifdef CONFIG_PM
3169 static int ixgbevf_resume(struct pci_dev *pdev)
3170 {
3171         struct ixgbevf_adapter *adapter = pci_get_drvdata(pdev);
3172         struct net_device *netdev = adapter->netdev;
3173         u32 err;
3174
3175         pci_set_power_state(pdev, PCI_D0);
3176         pci_restore_state(pdev);
3177         /*
3178          * pci_restore_state clears dev->state_saved so call
3179          * pci_save_state to restore it.
3180          */
3181         pci_save_state(pdev);
3182
3183         err = pci_enable_device_mem(pdev);
3184         if (err) {
3185                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3186                 return err;
3187         }
3188         pci_set_master(pdev);
3189
3190         rtnl_lock();
3191         err = ixgbevf_init_interrupt_scheme(adapter);
3192         rtnl_unlock();
3193         if (err) {
3194                 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3195                 return err;
3196         }
3197
3198         ixgbevf_reset(adapter);
3199
3200         if (netif_running(netdev)) {
3201                 err = ixgbevf_open(netdev);
3202                 if (err)
3203                         return err;
3204         }
3205
3206         netif_device_attach(netdev);
3207
3208         return err;
3209 }
3210
3211 #endif /* CONFIG_PM */
3212 static void ixgbevf_shutdown(struct pci_dev *pdev)
3213 {
3214         ixgbevf_suspend(pdev, PMSG_SUSPEND);
3215 }
3216
3217 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3218                                                 struct rtnl_link_stats64 *stats)
3219 {
3220         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3221         unsigned int start;
3222         u64 bytes, packets;
3223         const struct ixgbevf_ring *ring;
3224         int i;
3225
3226         ixgbevf_update_stats(adapter);
3227
3228         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3229
3230         for (i = 0; i < adapter->num_rx_queues; i++) {
3231                 ring = &adapter->rx_ring[i];
3232                 do {
3233                         start = u64_stats_fetch_begin_bh(&ring->syncp);
3234                         bytes = ring->total_bytes;
3235                         packets = ring->total_packets;
3236                 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3237                 stats->rx_bytes += bytes;
3238                 stats->rx_packets += packets;
3239         }
3240
3241         for (i = 0; i < adapter->num_tx_queues; i++) {
3242                 ring = &adapter->tx_ring[i];
3243                 do {
3244                         start = u64_stats_fetch_begin_bh(&ring->syncp);
3245                         bytes = ring->total_bytes;
3246                         packets = ring->total_packets;
3247                 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3248                 stats->tx_bytes += bytes;
3249                 stats->tx_packets += packets;
3250         }
3251
3252         return stats;
3253 }
3254
3255 static const struct net_device_ops ixgbevf_netdev_ops = {
3256         .ndo_open               = ixgbevf_open,
3257         .ndo_stop               = ixgbevf_close,
3258         .ndo_start_xmit         = ixgbevf_xmit_frame,
3259         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
3260         .ndo_get_stats64        = ixgbevf_get_stats,
3261         .ndo_validate_addr      = eth_validate_addr,
3262         .ndo_set_mac_address    = ixgbevf_set_mac,
3263         .ndo_change_mtu         = ixgbevf_change_mtu,
3264         .ndo_tx_timeout         = ixgbevf_tx_timeout,
3265         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
3266         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
3267 };
3268
3269 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3270 {
3271         dev->netdev_ops = &ixgbevf_netdev_ops;
3272         ixgbevf_set_ethtool_ops(dev);
3273         dev->watchdog_timeo = 5 * HZ;
3274 }
3275
3276 /**
3277  * ixgbevf_probe - Device Initialization Routine
3278  * @pdev: PCI device information struct
3279  * @ent: entry in ixgbevf_pci_tbl
3280  *
3281  * Returns 0 on success, negative on failure
3282  *
3283  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3284  * The OS initialization, configuring of the adapter private structure,
3285  * and a hardware reset occur.
3286  **/
3287 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3288 {
3289         struct net_device *netdev;
3290         struct ixgbevf_adapter *adapter = NULL;
3291         struct ixgbe_hw *hw = NULL;
3292         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3293         static int cards_found;
3294         int err, pci_using_dac;
3295
3296         err = pci_enable_device(pdev);
3297         if (err)
3298                 return err;
3299
3300         if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
3301             !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
3302                 pci_using_dac = 1;
3303         } else {
3304                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
3305                 if (err) {
3306                         err = dma_set_coherent_mask(&pdev->dev,
3307                                                     DMA_BIT_MASK(32));
3308                         if (err) {
3309                                 dev_err(&pdev->dev, "No usable DMA "
3310                                         "configuration, aborting\n");
3311                                 goto err_dma;
3312                         }
3313                 }
3314                 pci_using_dac = 0;
3315         }
3316
3317         err = pci_request_regions(pdev, ixgbevf_driver_name);
3318         if (err) {
3319                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3320                 goto err_pci_reg;
3321         }
3322
3323         pci_set_master(pdev);
3324
3325         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3326                                    MAX_TX_QUEUES);
3327         if (!netdev) {
3328                 err = -ENOMEM;
3329                 goto err_alloc_etherdev;
3330         }
3331
3332         SET_NETDEV_DEV(netdev, &pdev->dev);
3333
3334         pci_set_drvdata(pdev, netdev);
3335         adapter = netdev_priv(netdev);
3336
3337         adapter->netdev = netdev;
3338         adapter->pdev = pdev;
3339         hw = &adapter->hw;
3340         hw->back = adapter;
3341         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3342
3343         /*
3344          * call save state here in standalone driver because it relies on
3345          * adapter struct to exist, and needs to call netdev_priv
3346          */
3347         pci_save_state(pdev);
3348
3349         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3350                               pci_resource_len(pdev, 0));
3351         if (!hw->hw_addr) {
3352                 err = -EIO;
3353                 goto err_ioremap;
3354         }
3355
3356         ixgbevf_assign_netdev_ops(netdev);
3357
3358         adapter->bd_number = cards_found;
3359
3360         /* Setup hw api */
3361         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3362         hw->mac.type  = ii->mac;
3363
3364         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3365                sizeof(struct ixgbe_mbx_operations));
3366
3367         /* setup the private structure */
3368         err = ixgbevf_sw_init(adapter);
3369         if (err)
3370                 goto err_sw_init;
3371
3372         /* The HW MAC address was set and/or determined in sw_init */
3373         if (!is_valid_ether_addr(netdev->dev_addr)) {
3374                 pr_err("invalid MAC address\n");
3375                 err = -EIO;
3376                 goto err_sw_init;
3377         }
3378
3379         netdev->hw_features = NETIF_F_SG |
3380                            NETIF_F_IP_CSUM |
3381                            NETIF_F_IPV6_CSUM |
3382                            NETIF_F_TSO |
3383                            NETIF_F_TSO6 |
3384                            NETIF_F_RXCSUM;
3385
3386         netdev->features = netdev->hw_features |
3387                            NETIF_F_HW_VLAN_TX |
3388                            NETIF_F_HW_VLAN_RX |
3389                            NETIF_F_HW_VLAN_FILTER;
3390
3391         netdev->vlan_features |= NETIF_F_TSO;
3392         netdev->vlan_features |= NETIF_F_TSO6;
3393         netdev->vlan_features |= NETIF_F_IP_CSUM;
3394         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3395         netdev->vlan_features |= NETIF_F_SG;
3396
3397         if (pci_using_dac)
3398                 netdev->features |= NETIF_F_HIGHDMA;
3399
3400         netdev->priv_flags |= IFF_UNICAST_FLT;
3401
3402         init_timer(&adapter->watchdog_timer);
3403         adapter->watchdog_timer.function = ixgbevf_watchdog;
3404         adapter->watchdog_timer.data = (unsigned long)adapter;
3405
3406         INIT_WORK(&adapter->reset_task, ixgbevf_reset_task);
3407         INIT_WORK(&adapter->watchdog_task, ixgbevf_watchdog_task);
3408
3409         err = ixgbevf_init_interrupt_scheme(adapter);
3410         if (err)
3411                 goto err_sw_init;
3412
3413         strcpy(netdev->name, "eth%d");
3414
3415         err = register_netdev(netdev);
3416         if (err)
3417                 goto err_register;
3418
3419         netif_carrier_off(netdev);
3420
3421         ixgbevf_init_last_counter_stats(adapter);
3422
3423         /* print the MAC address */
3424         hw_dbg(hw, "%pM\n", netdev->dev_addr);
3425
3426         hw_dbg(hw, "MAC: %d\n", hw->mac.type);
3427
3428         hw_dbg(hw, "Intel(R) 82599 Virtual Function\n");
3429         cards_found++;
3430         return 0;
3431
3432 err_register:
3433         ixgbevf_clear_interrupt_scheme(adapter);
3434 err_sw_init:
3435         ixgbevf_reset_interrupt_capability(adapter);
3436         iounmap(hw->hw_addr);
3437 err_ioremap:
3438         free_netdev(netdev);
3439 err_alloc_etherdev:
3440         pci_release_regions(pdev);
3441 err_pci_reg:
3442 err_dma:
3443         pci_disable_device(pdev);
3444         return err;
3445 }
3446
3447 /**
3448  * ixgbevf_remove - Device Removal Routine
3449  * @pdev: PCI device information struct
3450  *
3451  * ixgbevf_remove is called by the PCI subsystem to alert the driver
3452  * that it should release a PCI device.  The could be caused by a
3453  * Hot-Plug event, or because the driver is going to be removed from
3454  * memory.
3455  **/
3456 static void ixgbevf_remove(struct pci_dev *pdev)
3457 {
3458         struct net_device *netdev = pci_get_drvdata(pdev);
3459         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3460
3461         set_bit(__IXGBEVF_DOWN, &adapter->state);
3462
3463         del_timer_sync(&adapter->watchdog_timer);
3464
3465         cancel_work_sync(&adapter->reset_task);
3466         cancel_work_sync(&adapter->watchdog_task);
3467
3468         if (netdev->reg_state == NETREG_REGISTERED)
3469                 unregister_netdev(netdev);
3470
3471         ixgbevf_clear_interrupt_scheme(adapter);
3472         ixgbevf_reset_interrupt_capability(adapter);
3473
3474         iounmap(adapter->hw.hw_addr);
3475         pci_release_regions(pdev);
3476
3477         hw_dbg(&adapter->hw, "Remove complete\n");
3478
3479         kfree(adapter->tx_ring);
3480         kfree(adapter->rx_ring);
3481
3482         free_netdev(netdev);
3483
3484         pci_disable_device(pdev);
3485 }
3486
3487 /**
3488  * ixgbevf_io_error_detected - called when PCI error is detected
3489  * @pdev: Pointer to PCI device
3490  * @state: The current pci connection state
3491  *
3492  * This function is called after a PCI bus error affecting
3493  * this device has been detected.
3494  */
3495 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
3496                                                   pci_channel_state_t state)
3497 {
3498         struct net_device *netdev = pci_get_drvdata(pdev);
3499         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3500
3501         netif_device_detach(netdev);
3502
3503         if (state == pci_channel_io_perm_failure)
3504                 return PCI_ERS_RESULT_DISCONNECT;
3505
3506         if (netif_running(netdev))
3507                 ixgbevf_down(adapter);
3508
3509         pci_disable_device(pdev);
3510
3511         /* Request a slot slot reset. */
3512         return PCI_ERS_RESULT_NEED_RESET;
3513 }
3514
3515 /**
3516  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
3517  * @pdev: Pointer to PCI device
3518  *
3519  * Restart the card from scratch, as if from a cold-boot. Implementation
3520  * resembles the first-half of the ixgbevf_resume routine.
3521  */
3522 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
3523 {
3524         struct net_device *netdev = pci_get_drvdata(pdev);
3525         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3526
3527         if (pci_enable_device_mem(pdev)) {
3528                 dev_err(&pdev->dev,
3529                         "Cannot re-enable PCI device after reset.\n");
3530                 return PCI_ERS_RESULT_DISCONNECT;
3531         }
3532
3533         pci_set_master(pdev);
3534
3535         ixgbevf_reset(adapter);
3536
3537         return PCI_ERS_RESULT_RECOVERED;
3538 }
3539
3540 /**
3541  * ixgbevf_io_resume - called when traffic can start flowing again.
3542  * @pdev: Pointer to PCI device
3543  *
3544  * This callback is called when the error recovery driver tells us that
3545  * its OK to resume normal operation. Implementation resembles the
3546  * second-half of the ixgbevf_resume routine.
3547  */
3548 static void ixgbevf_io_resume(struct pci_dev *pdev)
3549 {
3550         struct net_device *netdev = pci_get_drvdata(pdev);
3551         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3552
3553         if (netif_running(netdev))
3554                 ixgbevf_up(adapter);
3555
3556         netif_device_attach(netdev);
3557 }
3558
3559 /* PCI Error Recovery (ERS) */
3560 static const struct pci_error_handlers ixgbevf_err_handler = {
3561         .error_detected = ixgbevf_io_error_detected,
3562         .slot_reset = ixgbevf_io_slot_reset,
3563         .resume = ixgbevf_io_resume,
3564 };
3565
3566 static struct pci_driver ixgbevf_driver = {
3567         .name     = ixgbevf_driver_name,
3568         .id_table = ixgbevf_pci_tbl,
3569         .probe    = ixgbevf_probe,
3570         .remove   = ixgbevf_remove,
3571 #ifdef CONFIG_PM
3572         /* Power Management Hooks */
3573         .suspend  = ixgbevf_suspend,
3574         .resume   = ixgbevf_resume,
3575 #endif
3576         .shutdown = ixgbevf_shutdown,
3577         .err_handler = &ixgbevf_err_handler
3578 };
3579
3580 /**
3581  * ixgbevf_init_module - Driver Registration Routine
3582  *
3583  * ixgbevf_init_module is the first routine called when the driver is
3584  * loaded. All it does is register with the PCI subsystem.
3585  **/
3586 static int __init ixgbevf_init_module(void)
3587 {
3588         int ret;
3589         pr_info("%s - version %s\n", ixgbevf_driver_string,
3590                 ixgbevf_driver_version);
3591
3592         pr_info("%s\n", ixgbevf_copyright);
3593
3594         ret = pci_register_driver(&ixgbevf_driver);
3595         return ret;
3596 }
3597
3598 module_init(ixgbevf_init_module);
3599
3600 /**
3601  * ixgbevf_exit_module - Driver Exit Cleanup Routine
3602  *
3603  * ixgbevf_exit_module is called just before the driver is removed
3604  * from memory.
3605  **/
3606 static void __exit ixgbevf_exit_module(void)
3607 {
3608         pci_unregister_driver(&ixgbevf_driver);
3609 }
3610
3611 #ifdef DEBUG
3612 /**
3613  * ixgbevf_get_hw_dev_name - return device name string
3614  * used by hardware layer to print debugging information
3615  **/
3616 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
3617 {
3618         struct ixgbevf_adapter *adapter = hw->back;
3619         return adapter->netdev->name;
3620 }
3621
3622 #endif
3623 module_exit(ixgbevf_exit_module);
3624
3625 /* ixgbevf_main.c */